Valentin Boju Louis Funar
The Math Problems Notebook
Birkh¨auser Boston • Basel • Berlin
Valentin Boju MontrealTech...

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Valentin Boju Louis Funar

The Math Problems Notebook

Birkh¨auser Boston • Basel • Berlin

Valentin Boju MontrealTech Institut de Technologie de Montreal P.O. Box 78575, Station Wilderton Montreal, Quebec, H3S 2W9 Canada [email protected]

Louis Funar Institut Fourier BP 74 URF Mathematiques Universit´e de Grenoble I 38402 Saint Martin d’Heres cedex France [email protected]

Cover design by Alex Gerasev. Mathematics Subject Classification (2000): 00A07 (Primary); 05-01, 11-01, 26-01, 51-01, 52-01 Library of Congress Control Number: 2007929628 ISBN-13: 978-0-8176-4546-5

e-ISBN-13: 978-0-8176-4547-2

Printed on acid-free paper. c 2007 Birkh¨auser Boston All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Birkh¨auser Boston, c/o Springer Science+Business Media LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights.

9 8 7 6 5 4 3 2 1 www.birkhauser.com

(TXQ/SB)

The Authors

Valentin Boju was professor of mathematics at the University of Craiova, Romania, until his retirement in 2000. His research work was primarily in the field of geometry. He further promoted biostatistics and biomathematics as a discipline within the Department of Medicine of the University of Craiova. He left Romania for Canada, where he has taught at MontrealTech since 2001. He was actively involved in coaching college students in problem-solving and intuitive mathematics. In 2004, he was honored with the title of Officer of the Order “Cultural Merit,” Category “Scientific Research.” MontrealTech Institut de Technologie de Montréal P.O. Box 78575 Station Wilderton Montréal, Quebec, H3S 2W9 Canada e-mail: [email protected] Louis Funar has been a researcher at CNRS, at the Fourier Institute, University of Grenoble, France, since 1994. During his college years, he participated three times in the International Mathematics Olympiads with the Romanian team in the years 1983–1985, winning bronze, gold, and silver medals. His main research interests are in geometric topology and mathematical physics. Institut Fourier, BP 74 Mathématiques UMR 5582 Université de Grenoble I 38402 Saint-Martin-d’Hères cedex France e-mail: [email protected]

To all schoolteachers, particularly to my wonderful parents, Evdochia and Nicolae Boju, who genuinely believed in the education of the younger generation.

To my parents, Maria and Ioan Funar, who laid the foundation for our collaboration, as a late reward for their hard work and help, for their enthusiasm, determinedness, and strength during all these years.

Preface

The authors met on a Sunday morning about 25 years ago in Room 113. One of us was a college student and the other was leading the Sunday Math Circle. This circle, within the math department of the University of Craiova, gathered college students who possessed a common passion for mathematics. Most of them were participating in the mathematical competitions in vogue at that time, namely the Olympiads. Others just wanted to have good time. There were similar math circles everywhere in the country, in which high-school students were committed to active training for math competitions. The highest standard was achieved by the selective training camps organized at the national level, which were led by professional coaches and mathematicians who were able to stimulate and elicit high performance. To mention a few among the leaders, we recall Dorin Andrica, Toma Albu, Titu Andreescu, Mircea Becheanu, Ioan Cuculescu, Dorel Mihet, and Ioan Tomescu. The Sunday Math Circle shifted partially from its purpose of Olympiad training by following freely the leader’s ideas and thus becoming a place primarily intended for disseminating beautiful mathematics at an elementary level. The fundamental texts were the celebrated book of Richard Courant and Herbert Robbins with the mysterious title What is mathematics?, and the book of David Hilbert and Stefan Cohn–Vossen entitled Geometry and the Imagination. The participants soon realized the differences in both scale and novelty between the competition-type math problems that they encountered every day and the mathematics built up over hundreds of years and engaged in by professional mathematicians. Competition problems have to be solved in a short amount of time, and people compete against each other. These might be highly nontrivial, unconventional problems requiring deep insights and a lot of imagination, but still they have the advantage that one knows in advance that they have a solution. In the real mathematical world, problems often had to wait not merely years but sometimes centuries to be solved. A working mathematician could go for months or years trying to solve a particular problem that had not been solved before and take the risk of bitterly failing. Moreover, the sustained effort needed for accomplishing such a task would have to take into account a delicate balance between the accumulation of knowledge, methods, and tools and

viii

Preface

the creation from scratch of a path leading to a solution. This intellectual adventure is filled with suspense and frustration, since one does not know for sure what one is trying to prove or whether it is indeed true. Real mathematics seemed to many of us to be too far away from competition problems. The philosophy of the Sunday Math Circle was that, in contrast to what we might think, the border between the two kinds of mathematics could be so vague and flexible that at times one could cross it at an early age. The history of mathematics abounds in examples in which a fresh mind was able to find an unexpected solution that specialists had been unable to find. One needs, however, the right training and the unconventional sense of finding the inspiring problem. Eventually, once a solution has been found, mathematicians are then willing to try to understand it even better, and other solutions follow in time, each one simpler and clearer than the previous one. In some sense, once solved, even the hardest problems start losing slowly their aura of difficulty and eventually become just problems. Problems that are today part of the curriculum of the average high-school student were difficult research problems three hundred years ago, solved only by brilliant mathematicians. This phenomenon demonstrates the evolution that language and science have undergone since then. The authors conceived the present book with the nostalgia of the “good old times" of the Sunday Math Circles. We wanted something that carries the mark of that philosophy, namely, a number of challenging math problems for Olympiads with a glimpse of related problems of interest for the mathematician. The present text is a collection of problems that we think will be useful in training students for mathematical competitions. On the other hand, we hope that it might fulfill our second goal, namely, that of awakening interest in advanced mathematics. Thus its audience might range from college students and teachers to advanced math students and mathematicians. The problems in each section are in increasing order of difficulty, so that the reader give some of the problems a try. We wanted to have 25% easy problems concerning basic tools and methods and consisting mainly of instructional exercises. The beginner might jump directly to the solutions, where a concentrate of the general theory and basic tricks can be found. The largest chunk contains about 50% problems of medium difficulty, which could be useful in training for mathematical competitions from local to international levels. The remaining 25% might be considered difficult problems even for the experienced problem-solver. These problems are often accompanied by comments that put the results in a broader perspective and might incite the reader to pursue the research further. The problems serve as an excuse for introducing all sorts of generalizations and closely related open problems, which are spread among the solutions. Some of these are truly outstanding problems that resisted the efforts of mathematicians over the centuries, such as the congruent numbers conjecture and the Riemann hypothesis, which are among seven Millennium Prize Problems that the Clay Mathematics Institute recorded as some of the most difficult issues with which mathematicians were struggling at the turn of the second millennium and offered a reward of one million dollars for a solution to each one. In mathematics the frontier of our knowledge is still open, and it abounds in important unsolved problems, many of which can be

Preface

ix

understood at the undergraduate level. The reader might be soon driven to the edge of that part of mathematics where he or she could undertake original research. We drew inspiration from the spirit of the famous books by R. K. Guy and his collaborators. Nevertheless, our aim was not to build up a collection of open questions in elementary mathematics but rather to offer a journey among the basic methods in problem-solving. Developing intuition and strengthening the most popular techniques in mathematical competitions are equally part of the goal. In the meantime, we offer the enthusiastic reader a brief glimpse of mathematical research, a place where problems yet unsolved long for deliverance. The present collection of problems evolved from a notebook in which the secondnamed author collected the most interesting and unconventional problems that he encountered during his training for mathematical competitions in the 1980s. In the tradition of the Romanian school of mathematical training, he encountered problems inspired by both Russian Olympiads and American Competitions. Some (if not most) of the problems have already appeared elsewhere, especially in Kvant, Matematika v Shkole, American Mathematical Monthly, Elemente der Mathematik, Matematikai Lapok, Gazeta Matematica, and so, in a sense, the collection gained a certain cosmopolitan flavor. We have given the source in the problem section, when known, and more detailed information in the comments within the solutions part. The original set of problems is complemented by more basic exercises, which aim at introducing many of the tricks and methods of which the competitor should be aware. Years later, we followed the destiny of some of the most intriguing questions from the notebook. Some of them led to developments that are well beyond the scope of this book, and we decided to outline a few of them. We have supplied a short glossary containing some less usual definitions and identities in the geometry of triangles and the solution of Pell’s equation. In order to help the reader find his or her way through the book, we have provided both an index concerning all mathematical results needed in the proofs, which are usually stated at the place where they are used first, and an index of mathematical terms at the end of the book. The authors have benefited from discussions, corrections, help, and feedback from several people, whom we want to thank warmly: Dorin Andrica, Barbu Berceanu, Roland Bacher, Maxime Wolff, Mugurel Barcˇau, Ioan Filip, and Simon György Szatmari. We also thank Ann Kostant, Editorial Director, Springer, and Avanti Paranjpye, Associate Editor, Birkhäuser Boston, for their productive suggestions. One of them led to the “Index of Topics and Methods,” which might be useful for a better reception of the book by readers. We are very grateful to Elizabeth Loew, our Production Editor, for her patience and dedication to accuracy and excellence. Finally, we are thankful to David Kramer for his thorough copyediting corrections. Valentin Boju and Louis Funar Montreal, Canada and Saint-Martin-d’Hères, France July 2006

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Part I Problems 1

Number Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

2

Algebra and Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Algebraic Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Geometric Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11 11 13 18

3

Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Synthetic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Combinatorial Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Geometric Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21 21 23 27

4

Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Part II Solutions and Comments to the Problems 5

Number Theory Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

6

Algebra and Combinatorics Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Algebraic Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Geometric Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

Geometry Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 7.1 Synthetic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 7.2 Combinatorial Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

85 85 95 120

xii

Contents

7.3

Geometric Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8

Analysis Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

9

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 Compendium of Triangle Basic Terminology and Formulas . . . . . . . 9.1.1 Lengths in a Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1.2 Important Points and Lines in a Triangle . . . . . . . . . . . . . . . . . 9.1.3 Coordinates in a Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 Appendix: Pell’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

217 217 217 218 219 220

Index of Mathematical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 Index of Mathematical Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Index of Topics and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

Part I

Problems

1 Number Theory

Problem 1.1. Show that we have Cnk ≡ 0 (mod 2) for all k satisfying 1 ≤ k ≤ n − 1 if and only if n = 2β , where β ∈ Z∗+ . Here Cnk denotes the number of combinations, i.e., the number of ways of picking up a subset of k elements from a set of n elements. Known also as the binomial coefficient or choice number and sometimes denoted as n it is given by the formula k n n! Cnk = = k k!(n − k)! where the factorial n! represents n! = 1 × 2 × 3 × · · · × n. Problem 1.2. Let P = an x n + · · · + a0 be a polynomial with integer coefficients. Suppose that there exists a number p such that: 1. p does not divide an ; 2. p divides ai , for all i ≤ n − 1; 3. p 2 does not divide a0 . Then P is an irreducible polynomial in Z[x]. Problem 1.3. Given mi , bi ∈ Z+ , i ∈ {1, 2, . . . , n}, such that gcd(mi , mj ) = 1 for all i = j , there exist integers x satisfying x ≡ bi (mod mi ) for all i. This result is usually known as the Chinese remainder theorem. Problem 1.4. If gcd(a, m) = 1, then it is a classical result of Euler that we have the following congruence: a ϕ(m)+1 ≡ 0 (mod m), where ϕ(m) is the Euler totient function, which counts how many positive integers smaller than m are relatively prime to m. Prove that this equality holds precisely for those numbers a, m such that for any prime number p that divides a, if pk divides m then pk also divides a.

4

1 Number Theory

Problem 1.5. (Kvant) Let p > 2 be a prime number and ak ∈ {0, 1, . . . , p2 − 1} denote the value of k p modulo p 2 . Prove that p−1

ak =

k=1

p3 − p2 . 2

Problem 1.6. If k ∈ Z+ , then show that k 2 + 1 + · · · + k 2 + 2k = 2k 2 + 2k, where [x] denotes the integer part, i.e., the largest integer smaller than x. Problem 1.7. (Amer. Math. Monthly) If n is not a multiple of 5, then P = x 4n + x 3n + x 2n + x n + 1 is divisible by Q = x 4 + x 3 + x 2 + x + 1. Problem 1.8. (Amer. Math. Monthly) Let p be an odd prime and k ∈ Z+ . Show that there exists a perfect square the last k digits of whose expansion in base p are 1. Problem 1.9. Any natural number greater than 6 can be written as a sum of two numbers that are relatively prime. Problem 1.10. (International Math. Olympiad) Prove that it is impossible to extract an infinite arithmetic progression from the sequence S = {1, 2k , 3k , . . . , nk , . . . }, where k ≥ 2. Problem 1.11. (Amer. Math. Monthly) Prove that ba−j +1 divides Cba if a, b ≥ 2, j ≤ a + 1. j

Problem 1.12. Solve in integers the following equations: (1) x 2 = y 2 + y 3 ; (2) x 2 + y 2 = z2 . Problem 1.13. Let a, b, c, d ∈ Z+ be such that at least one of a and c is not a perfect square and gcd(a, c) = 1. Show that there exist infinitely many natural numbers n such that an + b, cn + d are simultaneously perfect squares if one of the following conditions is satisfied: 1. b and d are perfect squares; 2. a + b, c + d are perfect squares; 3. a(d − 1) = c(b − 1). Moreover, there do not exist such numbers if a = 1, b = 0, c = 4k 2 − 1, d = 1. √ Problem 1.14. If N = 2 + 2 28n2 + 1 ∈ Z for a natural number n, then N is a perfect square. Problem 1.15. Let n ≥ 5, 2 ≤ b ≤ n. Prove that

(n − 1)! ≡ 0 (mod b − 1). b

1 Number Theory

5

Problem 1.16. (Amer. Math. Monthly) Prove that for every natural number n, there exists a natural number k such that k appears in exactly n nontrivial Pythagorean triples. Problem 1.17. (Amer. Math. Monthly) Let n, q ∈ Z+ be such that all prime divisors of q are greater than n. Show that (q − 1)(q 2 − 1) · · · (q n−1 − 1) ≡ 0 (mod n!). Problem 1.18. Every natural number n ≥ 6 can be written as a sum of distinct primes. Problem 1.19. Every number n ≥ 6 can be written as a sum of three numbers that are pairwise relatively prime. Problem 1.20. (Romanian training camp, Sinaia 1984) Find all pairs of integers (m, n) such that n Cm = 1984, n = where Cm

m! n!(m−n)!

denotes the usual binomial coefficient.

Problem 1.21. (Romanian training camp, Sinaia 1984) Find the set A consisting of natural numbers n that are divisible by all odd natural numbers a with a 2 < n. Problem 1.22. Prove that 21092 − 1 is divisible by 10932 . Problem 1.23. (Amer. Math. Monthly) Let n ≥ 0, r > 1, and 0 < a ≤ r be three integers. Prove that the number n, when written in base r, has precisely ∞

nr −k + ar −1 − nr −k

k=1

digits that are greater than or equal to r − a. Problem 1.24. Find a pair (a, b) of natural numbers satisfying the following properties: 1. ab(a + b) is not divisible by 7. 2. (a + b)7 − a 7 − b7 is divisible by 77 . Problem 1.25. (International Math. Olympiad 1984) Let 0 < a < b < c < d be odd integers such that 1. ad = bc 2. a + d = 2k , b + c = 2m , for some integers k and m. Prove that a = 1. Problem 1.26. (Amer. Math. Monthly) Find those subsets S ⊂ Z+ such that all but finitely many sums of elements from S (possibly with repetitions) are composite numbers.

6

1 Number Theory

Problem 1.27. (Amer. Math. Monthly) Prove that for any natural n > 1, the number 2n − 1 does not divide 3n − 1. Problem 1.28. (Putnam Competition 1964) Let un be the least common multiple of the first n terms of a strictly increasing sequence of positive integers. Prove that ∞ 1 ≤ 2. un n=1

Find a sequence for which equality holds above. Problem 1.29. (Amer. Math. Monthly) Let ϕn (m) = ϕ(ϕn−1 (m)), where ϕ1 (m) = ϕ(m) is the Euler totient function, and set ω(m) the smallest number n such that ϕn (m) = 1. If m < 2α , then prove that ω(m) ≤ α. Problem 1.30. Let f be a polynomial with integer coefficients and N (f ) = card{k ∈ Z; f (k) = ±1}. Prove that N (f ) ≤ 2 + deg f , where deg f denotes the degree of f . Problem 1.31. Prove that every integer can be written as a sum of 5 perfect cubes. Problem 1.32. If n ∈ Z, then the binomial coefficient C2n = number of divisors.

(2n)! (n!)2

has an even

Problem 1.33. (Amer. Math. Monthly) Prove that every n ∈ Z+ can be written in precisely k(n) different ways as a sum of consecutive integers, where k(n) is the number of odd divisors of n greater than 1. Problem 1.34. (Amer. Math. Monthly) Let π2 (x) denote the number of twin primes p with p ≤ x. Recall that p is a twin prime if both p and p + 2 are prime. Show that

π n! π (n − 2)! sin (n + 2) sin n π2 (x) = 2 + 2 n+2 2 n 7≤n≤x

for x > 7. Problem 1.35. (Kvant) 1. Find all solutions of the equation 3x+1 + 100 = 7x−1 . 2 2 2. Find two solutions of the equation 3x + 3x = 2x + 4x , and prove that there are no others. Problem 1.36. (Kvant) Let σ (n) denote the sum of the divisors of n. Prove that there exist infinitely many integers n such that σ (n) > 2n, or even stronger, such that σ (n) > 3n. Prove also that σ (n) < n(1 + log n). Problem 1.37. (American Competition) Let ai be natural numbers such that gcd(ai , aj ) = 1 and the ai are not prime numbers. Show that 1 1 + ··· + < 2. a1 an

1 Number Theory

7

Problem 1.38. Let σ (n) denote the sum of divisors of n. Show that σ (n) = 2k if and only if n is a product of Mersenne primes, i.e., primes of the form 2k − 1, for k ∈ Z. Problem 1.39. (Putnam Competition) Find all integer solutions of the equation |pr − q s | = 1, where p, q are primes and r, s ∈ Z \ {0, 1}. Problem 1.40. Consider an arithmetic progression with ratio between 1 and 2000. Show that the progression does not contain more than 10 consecutive primes. √ Problem 1.41. (Amer. Math. Monthly) Let a1 = 1, an+1 = an + an . Show that an is a perfect square iff n is of the form 2k + k − 2. Problem 1.42. Recall that ϕ(n) denotes the Euler totient function (i.e., the number of natural numbers less than n and prime to n), and that σ (n) is the sum of divisors of n. Show that n is prime iff ϕ(n) divides n − 1 and n + 1 divides σ (n). Problem 1.43. (Amer. Math. Monthly) A number is called ϕ-subadditive if ϕ(n) ≤ ϕ(k) + ϕ(n − k) for all k such that 1 ≤ k ≤ n − 1, and ϕ-superadditive if the reverse inequality holds. Prove that there are infinitely many ϕ-subadditive numbers and infinitely many ϕ-superadditive numbers. Problem 1.44. (Amer. Math. Monthly) Find the positive integers N such that for all n ≥ N, we have ϕ(n) ≤ ϕ(N ). Problem 1.45. A number n is perfect if σ (n) = 2n, where σ (n) denotes the sum of all divisors of n. Prove that the even number n is perfect if and only if n = 2p−1 (2p − 1), where p is a prime number with the property that 2p − 1 is prime. Problem 1.46. (Elemente der Mathematik) A number n is superperfect if σ (σ (n)) = 2n, where σ (k) is the sum of all divisors of k. Prove that the even number n is superperfect if and only if n = 2r , where r is an integer such that 2r+1 − 1 is prime. Problem 1.47. If a, b are rational numbers satisfying tan aπ = b, then b ∈ {−1, 0, 1}. Problem 1.48. (Amer. Math. Monthly) Let An = run + sv n , n ∈ Z+ , where r, s, u, v are integers, u = ±v, and let Pn be the set of prime divisors of An . Then P = ∞ n=0 Pn is infinite. Problem 1.49. Solve in natural numbers the equation x 2 + y 2 + z2 = 2xyz. Problem 1.50. (Amer. Math. Monthly) Find the greatest common divisor of the fol1 , C 3 , C 5 , . . . , C 2n−1 . lowing numbers: C2n 2n 2n 2n Problem 1.51. (Amer. Math. Monthly) Prove that if Sk = ni=1 k gcd(i,n) , then Sk ≡ 0 (mod n).

8

1 Number Theory

Problem 1.52. Prove that for any integer n such that 4 ≤ n ≤ 1000, the equation 1 1 1 4 + + = x y z n has solutions in natural numbers. Problem 1.53. (Nieuw Archief v. Wiskunde) For a natural number n we set S(n) for the set of integers that can be written in the form 1 + g + · · · + g n−1 for some g ∈ Z+ , g ≥ 2. 1. Prove that S(3) ∩ S(4) = ∅. 2. Find S(3) ∩ S(5). Problem 1.54. If k ≥ 202 and n ≥ 2k, then prove that Cnk > nπ(k) , where π(k) denotes the number of prime numbers smaller than k. Problem 1.55. (Amer. Math. Monthly) Let Sm (n) be the sum of the inverses of the integers smaller than m and relatively prime to n. If m > n ≥ 2, then show that Sm (n) is not an integer. Problem 1.56. Solve in integers the following equations: 1. x 4 + y 4 = z2 ; 2. y 3 + 4y = z2 ; 3. x 4 = y 4 + z2 . Problem 1.57. Show that the equation x 3 + y 3 = z3 + w 3 has infinitely many integer solutions. Prove that 1 can be written in infinitely many ways as a sum of three cubes. Problem 1.58. Prove that the equation y 2 = Dx 4 + 1 has no integer solution except x = 0 if D ≡ 0, −1, 3, 8 (mod 16) and there is no factorization D = pq, where p > 1 is odd, gcd(p, q) = 1, and either p ≡ ±1 (mod 16), p ≡ q ± 1 (mod 16), or p ≡ 4q ± 1 (mod 16). Problem 1.59. (Amer. Math. Monthly) Find all numbers that are simultaneously triangular, perfect squares, and pentagonal numbers. Problem 1.60. (Amer. Math. Monthly) Find all inscribable integer-sided quadrilaterals whose areas equal their perimeters. Problem 1.61. (Amer. Math. Monthly) Every rational number a can be written as a sum of the cubes of three rational numbers. Moreover, if a > 0, then the three cubes can be chosen to be positive. Problem 1.62. 1. Every prime number of the form 4m + 1 can be written as the sum of two perfect squares.

1 Number Theory

9

2. Every natural number is the sum of four perfect squares. Problem 1.63. Every natural number can be written as a sum of at most 53 integers to the fourth power. Problem 1.64. Let G(k) denote the minimal integer n such that any positive integer k cank be written as the sum of n positive perfect kth powers. Prove that G(k) ≥ 2 + 3 − 2. 2k Problem 1.65. (Amer. Math. Monthly) Let R(n, k) be the remainder when n is divided by k and k R(n, i). S(n, k) = i=1 2

n) = 1 − π12 . 1. Prove that limn→∞ S(n, n2 2. Consider a sequence of natural numbers (ak ) growing to infinity and such that k limk→∞ ak log = 0. Prove that limk→∞ S(kakk2, k) = 41 . k

Problem 1.66. (Nieuw Archief v. Wiskunde) Let a1 < a2 < a3 < · · · < an < · · · be a sequence of positive integers such that the series ∞ 1 ai i=1

converges. Prove that for any i, there exist infinitely many sets of ai consecutive integers that are not divisible by aj for all j > i. Problem 1.67. (Amer. Math. Monthly) Denote by Cn the claim that there exists a set of n consecutive integers such that no two of them are relatively prime. Prove that Cn is true for every n such that 17 ≤ n ≤ 10000. Problem 1.68. (Schweitzer Competition) Prove that a natural number has more divisors that can be written in the form 3k + 1, for k ∈ Z, than divisors of the form 3m − 1, for m ∈ Z. Problem 1.69. (Amer. Math. Monthly) A number N is called deficient if σ (N ) < 2N and abundant if σ (N) > 2N . 1. Let k be fixed. Are there any sequences of k consecutive abundant numbers? 2. Show that there are infinitely many 5-tuples of consecutive deficient numbers. Problem 1.70. (Amer. Math. Monthly) Does there exist a nonconstant polynomial an2 +bn+c with integer coefficients such that for any natural numbers m all its prime factors pi are congruent to 3 modulo 4? Prove that for any nonconstant polynomial f with integer coefficients and any m ∈ Z there exist a prime number p and a natural number n such that p divides f (n) and p ≡ 1 (mod m).

2 Algebra and Combinatorics

2.1 Algebra p−1 Problem 2.1. Set Sk,p = i=1 i k , for natural numbers p and k. If p ≥ 3 is prime and 1 < k ≤ p − 2, show that Sk,p ≡ 0 (mod p). Problem 2.2. Let P = a0 +· · ·+an x n and Q = b0 +· · ·+bm x m be two polynomials with m ≤ n. Then, deg gcd(P , Q) ≥ 1 if and only if there exist two polynomials K and L, such that deg K ≤ m − 1, deg L ≤ n − 1, and K · P = L · Q. Prove that this is equivalent to the vanishing of the following (n + m) × (n + m) determinant: ⎛ ⎞ a0 a1 a2 · · · am−1 · · · an−1 an 0 0 ··· 0 ⎜ 0 a0 a1 · · · am−2 · · · an−2 an−1 an 0 ··· 0 ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. .. .. .. .. ⎟ ⎜ . . . ⎟ . . . . . . ⎜ ⎟ ⎜ 0 0 0 · · · a0 · · · an−m−1 an−m an−m+1 · · · a n ⎟ ⎜ ⎟ ⎜ 0 0 0 ··· 0 · · · an−m−2 an−m−1 an−m ··· an−1 ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. ⎟ = 0. det ⎜ ... ... ... . . . . . ⎟ ⎜ ⎟ ⎜ 0 0 0 ··· 0 0 a0 a1 a2 ··· am ⎟ ⎜ ⎟ ⎜ b0 b1 · · · bm−1 bm 0 0 0 0 ··· 0 ⎟ ⎜ ⎟ ⎜ 0 b0 · · · bm−2 bm−1 bm 0 0 0 ··· 0 ⎟ ⎜ ⎟ ⎜ . . .. .. .. .. .. .. .. ⎟ ⎝ .. .. . . . . . . . ⎠ 0 0 · · · b0 b1 · · · · · · ··· 0 ··· 0 Problem 2.3. Prove that if P , Q ∈ R[x, y] are relatively prime polynomials, then the system of equations P (x, y) = 0, Q(x, y) = 0, has only finitely many real solutions.

12

2 Algebra and Combinatorics

Problem 2.4. Let a, b, c, d ∈ R[x] be polynomials with real coefficients. Set x x x x p= ac dt, q = ad dt, r = bc dt, s = bd dt. 1

1

1

1

Prove that ps − qr is divisible by (x − 1)4 . Problem 2.5. 1. Find the minimum number of elements that must be deleted from the set {1, . . . , 2005} such that the set of the remaining elements does not contain two elements together with their product. 2. Does there exist, for any k, an arithmetic progression with k terms in the infinite sequence 1 1 1 1, , . . . , , . . . , , . . .? 2 2005 n Problem 2.6. (Amer. Math. Monthly) Consider a set S of n elements and n+1 subsets M1 , . . . , Mn+1 ⊂ S. Show that there exist r, s ≥ 1 and disjoint sets of indices {i1 , . . . , ir } ∩ {j1 , . . . , js } = ∅ such that r

Mik =

k=1

s

Mjk .

k=1

Problem 2.7. Let p be a prime number, and A = {a1 , . . . , ap−1 } ⊂ Z∗+ a set of integers that are not divisible by p. Define the map f : P(A) → {0, 1, . . . , p − 1} by k f ({ai1 , . . . , aik }) = aip (mod p), and f (∅) = 0. p−1

Prove that f is surjective. Problem 2.8. Consider the function Fr = x r sin rA + y r sin rB + zr sin rC, where x, y, z ∈ R, A + B + C = kπ, and r ∈ Z+ . Prove that if F1 (x0 , y0 , z0 ) = F2 (x0 , y0 , z0 ) = 0, then Fr (x0 , y0 , z0 ) = 0, for all r ∈ Z+ . Problem 2.9. Let T (z) ∈ Z[z] be a nonzero polynomial with the property that |T (ui )| ≤ 1 for all values ui that are roots of P (z) = zn − 1. Prove that either T (z) is divisible by P (z), or else there exists some k ∈ Z+ , k ≤ n − 1, such that T (z) ± zk is divisible by P (z). The same result holds when instead of P (z), we consider zn + 1. Problem 2.10. 1. If the map x → x 3 from a group G to itself is an injective group homomorphism, then G is an abelian. 2. If the map x → x 3 from a group G to itself is a surjective group homomorphism, then G is an abelian. 3. Find an abelian group with the property that x → x 4 is an automorphism. 4. What can be said for exponents greater than 4?

2.2 Algebraic Combinatorics

13

Problem 2.11. (Amer. Math. Monthly) Let V be a vector space of dimension n > 0 over a field of characteristic p = 0 and let A be an affine map A : V → V . Prove that there exist u ∈ V and 0 ≤ k ≤ np such that Ak u = u. Problem 2.12. (Putnam Competition 1959) Find the cubic equation the zeros of which are the cubes of the roots of the equation x 3 + ax 2 + bx + c = 0. Problem 2.13. (Putnam Competition 1956) Assume that the polynomials P , Q ∈ C[x] have the same roots, possibly with different multiplicities. Suppose, moreover, that the same holds true for the pair P + 1 and Q + 1. Prove that P = Q. Problem 2.14. (Amer. Math. Monthly) Determine r ∈ Q, for which 1, cos 2π r, sin 2π r are linearly dependent over Q. Problem 2.15. (Amer. Math. Monthly) 1. Prove that there exist a, b, c ∈ Z, not all zero, such that |a|, |b|, |c| < 106 √ √ a + b 2 + c 3 < 10−11 . 2. Prove that if 0 ≤ |a|, |b|, |c| 10−21 . Problem 2.16. (Amer. Math. Monthly) Prove that if n > 2, then we do not have any nontrivial solutions for the equation x n + y n = zn , where x, y, z are rational functions. Solutions of the form x = af, y = bf, z = cf , where f is a rational function and a, b, c are complex numbers satisfying a n +bn = cn , are called trivial. Problem 2.17. (Kvant) A table is an n × k rectangular grid drawn on the torus, every box being assigned an element from Z/2Z. We define a transformation acting on tables as follows. We replace all elements of the grid simultaneously, each element being changed into the sum of the numbers previously assigned to its neighboring boxes. Prove that iterating this transformation sufficiently many times, we always obtain the trivial table filled with zeros, no matter what the initial table was, if and only if n = 2p and k = 2q for some integers p, q. In this case we say that the respective n × k grid is nilpotent.

2.2 Algebraic Combinatorics Problem 2.18. Let us consider a four-digit number N whose digits are not all equal. We first arrange its digits in increasing order, then in decreasing order, and finally, we subtract the two obtained numbers. Let T (N) denote the positive difference thus obtained. Show that after finitely many iterations of the transformation T , we obtain 6174.

14

2 Algebra and Combinatorics

Problem 2.19. Find an example of a sequence of natural numbers 1 ≤ a1 < a2 < · · · < an < an+1 < · · · with the property that every m ∈ Z+ can be uniquely written as m = ai − aj , for i, j ∈ Z+ . Problem 2.20. (Amer. Math. Monthly) Consider the set of 2n integers {±a1 , ±a2 , . . . , ±an } and m < 2n . Show that we can choose a subset S such that 1. The two numbers ±ai are not both in S; 2. The sum of all elements of S is divisible by m. Problem 2.21. (Proposed for the International Math. Olympiad) Show that for every natural number n there exist prime numbers p and q such that n divides their difference. Problem 2.22. An even number, 2n, of knights arrive at King Arthur’s court, each one of them having at most n − 1 enemies. Prove that Merlin the wizard can assign places for them at a round table in such a way that every knight is sitting only next to friends. Problem 2.23. (Putnam Competition) Let r, s ∈ Z+ . Find the number of 4-tuples of positive integers (a, b, c, d) that satisfy 3r 7s = lcm(a, b, c) = lcm(a, b, d) = lcm(a, c, d) = lcm(b, c, d). Problem 2.24. (Putnam Competition) 1. Let n ∈ Z+ and p be a prime number. Denote by N (n, p) the number of binomial coefficients Cns that are not divisible by p. Assume that n is written in base p as n = n0 + n1 p + · · · + nm p m , where 0 ≤ nj < p for all j ∈ {0, 1, . . . , m}. Prove that N (n, p) = (n0 + 1)(n1 + 1) · · · (nj + 1). 2. Write k in base p as k = k0 + k1 p + · · · + kj p j , with 0 ≤ kj ≤ p − 1, for all j ∈ {0, 1, . . . , s}. Prove that k

Cnk ≡ Cnk00 Cnk11 · · · Cnjj (mod p). Problem 2.25. (Putnam Competition) Define the sequence Tn by T1 = 2, Tn+1 = Tn2 − Tn + 1, for n ≥ 1. Prove that if m = n, then Tm and Tn are relatively prime, and further, that ∞ 1 = 1. Ti i=1

Problem 2.26. Let α, β > 0 and consider the sequences [α], [2α], . . . , [kα], . . . ; [β], [2β], . . . , [kβ], . . . , where the brackets denote the integer part. Prove that these two sequences taken together enumerate Z+ in an injective manner if and only if α, β ∈ R \ Q and

1 1 + = 1. α β

2.2 Algebraic Combinatorics

15

Problem 2.27. We say that the sets S1 , S2 , . . . , Sm form a complementary system if they make a partition of Z+ , i.e., every positive integer belongs to a unique set Si . Let m > 1 and α1 , . . . , αm ∈ R+ . Then the sets Si = {[nαi ], where n ∈ Z+ } form a complementary system only if m = 2, α1−1 + α2−1 = 1, and α1 ∈ R \ Q. Problem 2.28. Let f : Z+ → Z+ be an increasing function and set F (n) = f (n) + n, G(n) = f ∗ (n) + n, where f ∗ (n) = card({x ∈ Z+ ; 0 ≤ f (x) < n}). Then {F (n); n ∈ Z+ } and {G(n); n ∈ Z+ } are complementary sequences. Conversely, any two complementary sequences can be obtained this way using some nondecreasing function f . Problem 2.29. Let M denote the set of bijective functions f : Z+ → Z+ . Prove that there is no bijective function between M and Z. Problem 2.30. (Amer. Math. Monthly) Let F ⊂ Z be a finite set of integers satisfying the following properties: 1. For any x ∈ F there exist y, z ∈ F such that x = y + z. 2. There exists n such that for any natural number 1 ≤ k ≤ n, and any choice of x1 , . . . , xk ∈ F , their sum x1 + · · · + xk is nonzero. Prove that card(F ) ≥ 2n + 2. Problem 2.31. (Amer. Math. Monthly) For a finite graph G we denote by Z(G) the minimal number of colors needed to color all its vertices such that adjacent vertices have different colors. This is also called the chromatic number of G. Prove that the inequality Z(G) ≥

p2

p2 − 2q

holds if G has p vertices and q edges. Problem 2.32. Let Dk be a collection of subsets of the set {1, . . . , n} with the property that whenever A = B ∈ Dk , then card(A ∩ B) ≤ k, where 0 ≤ k ≤ n − 1. Prove that card(Dk ) ≤ Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . Problem 2.33. Prove that

⎧ 0, ⎪ ⎪ ⎪ ⎪ 1, ⎪ ⎪ ⎪ ⎨ n/2,

if if n if 1 (−1)n−k Cnk k p = n(3n+1) , if ⎪ p! 24 ⎪ k=0 ⎪ n2 (n+1) ⎪ ⎪ , if ⎪ 48 ⎪ ⎩ n(15n3 +30n2 +5n+1) , if 1152

0 ≤ p < n, p = n, p = n + 1, p = n + 2, p = n + 3, p = n + 4.

16

2 Algebra and Combinatorics

Problem 2.34. Write lcm(a1 , . . . , an ) in terms of the various gcd(ai , . . . , aj ) for subsets of {a1 , . . . , an }. Problem 2.35. Let f (n) be the number of ways in which a convex polygon with n+1 sides can be divided into regions delimited by several diagonals that do not intersect (except possibly at their endpoints). We consider as distinct the dissection in which we first cut the diagonal a and next the diagonal b from the dissection in which we first cut the diagonal b and next the diagonal a. It is easy to compute the first values of f (n), as follows: f (1) = 1, f (2) = 1, f (3) = 3, f (4) = 11, f (5) = 45. Find the generating function F (x) = f (n)x n and an asymptotic formula for f (n). Problem 2.36. Find the permutation σ : (1, . . . , n) → (1, . . . , n) such that S(σ ) =

n

|σ (i) − i|

i=1

is maximal. Problem 2.37. (Amer. Math. Monthly) On the set Sn of permutations of {1, . . . , n} we define an invariant distance function by means of the formula d(σ, τ ) =

n

|σ (i) − τ (i)|.

i=1

What are the values that d could possibly take? Problem 2.38. (Preliminaries for the International Math. Olympiad, Romania) The set M = {1, 2, . . . , 2n} is partitioned into k sets M1 , . . . , Mk , where n ≥ k 3 + k. Show that there exist i, j ∈ {1, . . . , k} for which we can find k + 1 distinct even numbers 2r1 , . . . , 2rk+1 ∈ Mi with the property that 2r1 − 1, . . . , 2rk+1 − 1 ∈ Mj . Problem 2.39. (Preliminaries for the International Math. Olympiad, Great Britain) Let S be the set of odd integers not divisible by 5 and smaller than 30m, where m ∈ Z∗+ . Find the smallest k such that every subset A ⊂ S of k elements contains two distinct integers, one of which divides the other. ai −aj Problem 2.40. Prove that 1≤j nn−1 . Show that there exist distinct prime numbers p1 , p2 , . . . , pn such that C + j is divisible by pj . Problem 2.47. (Nieuw Archief v. Wiskunde) Let p be a prime number and f (p) the smallest integer for which there exists a partition of the set {2, 3, . . . , p} into f (p) classes such that whenever a1 , . . . , ak belong to the same class of the partition, the equation k xi ai = p i=1

does not have solutions in nonnegative integers. Estimate f (p). Problem 2.48. (Schweitzer Competition) Consider m distinct natural numbers ai √ smaller than N such that lcm(ai , aj ) ≤ N for all i, j . Prove that m ≤ 2 N . Problem 2.49. The set M ⊂ Z+ is called A-sum-free, where A = (a1 , a2 , . . . , ak ) ∈ Zk+ , if for any choice of x1 , x2 , . . . , xk ∈ M we have a1 x1 + a2 x2 + · · · + ak xk ∈ M. If A, B are two vectors, we define f (n; A, B) as the greatest number h such that there exists a partition of the set of consecutive integers {n, n + 1, . . . , h} into S1 and S2 such that S1 is A-sum-free and S2 is B-sum-free. Assume that B = (b1 , b2 , . . . , bm ) and that the conditions below are satisfied: a1 + a2 + · · · + ak = b1 + b2 + · · · + bm = s, and min aj = min bj = 1, k, m ≥ 2.

1≤j ≤k

Prove that f (n; A, B) =

ns 2

1≤j ≤m

+ n(s − 1) − 1.

Problem 2.50. (Amer. Math. Monthly) Let 1 ≤ a1 < a2 < · · · < an < 2n be a sequence of natural numbers for n ≥ 6. Prove that n min lcm(ai , aj ) ≤ 6 +1 . i,j 2 Moreover, the constant 6 is sharp.

18

2 Algebra and Combinatorics

Problem 2.51. (Amer. Math. Monthly) Let 1 ≤ a1 < a2 < · · · < ak < n be such that gcd(ai , aj ) = 1 for all 1 ≤ i < j ≤ k. Determine the maximum value of k. Problem 2.52. (International Math. Olympiad 1978) Consider the increasing sequence f (n) ∈ Z+ , 0 < f (1) < f (2) < · · · < f (n) < · · · . It is known that the nth element in increasing order among the positive integers that are not terms of this sequence is f (f (n)) + 1. Find the value of f (240). Problem 2.53. (Amer. Math. Monthly) We define inductively three sequences of integers (an ), (bn ), (cn ) as follows: 1. a1 = 1, b1 = 2, c1 = 4; 2. an is the smallest integer that does not belong to the set {a1 , . . . , an−1 , b1 , . . . , bn−1 , c1 , . . . , cn−1 }; 3. bn is the smallest integer that does not belong to the set {a1 , . . . , an−1 , an , b1 , . . . , bn−1 , c1 , . . . , cn−1 }; 4. cn = 2bn + n − an . Prove that

√ 0 < n 1 + 3 − bn < 2

for all n ∈ Z+ .

2.3 Geometric Combinatorics Problem 2.54. We consider n points in the plane that determine Cn2 segments, and to each segment one associates either +1 or −1. A triangle whose vertices are among these points will be called negative if the product of numbers associated to its sides is negative. Show that if n is even, then the number of negative triangles is even. Moreover, for odd n, the number of negative triangles has the same parity as the number p of segments labeled −1. Problem 2.55. (Amer. Math. Monthly) Given n find a finite set S consisting of natural numbers larger than n with the property that for any k ≥ n the k × k square can be tiled by a family of si × si squares, where si ∈ S. Problem 2.56. We consider 3n points A1 , . . . , A3n in the plane whose positions are defined recursively by means of the following rule: first, the triangle A1 A2 A3 is equilateral; further, the points A3k+1 , A3k+2 , and A3k+3 are the midpoints of the sides of the triangle A3k A3k−1 A3k−2 . Let us assume that the 3n points are colored with two colors. Show that for n ≥ 7 there exists at least one isosceles trapezoid having vertices of the same color. Problem 2.57. Is there a coloring of all lattice points in the plane using only two colors such that there are no! rectangles with all vertices of the same color whose side ratio belongs to 1, 21 , 13 , 23 ?

2.3 Geometric Combinatorics

19

Problem 2.58. (Amer. Math. Monthly) Let G be a planar graph and let P be a path in G. We say that P has a (transversal) self-intersection in the vertex v if the path has a (transversal) self-intersection from the curve-theoretic viewpoint. Let us give an example: Take the point 0 in the plane and the segments 01, 02, 03, 04 going counterclockwise around 0. Then a path traversing first 103 and then 204 has a (transversal) self-intersection at 0, while a path going first along 102 and further on 304 does not have a (transversal) self-intersection. Prove that any connected planar graph G with only even-degree vertices admits an Eulerian circuit without self-intersections. Recall that an Eulerian circuit is a path along the edges of the graph that passes precisely once along each edge of the graph. Problem 2.59. (Hungarian Competition) Let us consider finitely many points in the plane that are not all collinear. Assume that one associates to each point a number from the set {−1, 0, 1} such that the following property holds: for any line determined by two points from the set, the sum of numbers associated to all points lying on that line equals zero. Show that, if the number of points is at least three, then to each point one is associated with 0. Problem 2.60. (Amer. Math. Monthly) If one has a set of squares with √ total area smaller than 1, then one can arrange them inside a square of side length 2, without any overlaps. Problem 2.61. Prove that for each k there exist k points in the plane, no three collinear and having integral distance from each other. If we have an infinite set of points with integral distances from each other, then all points are collinear. Problem 2.62. (International Math. Olympiad 1984) Let O, A be distinct points in (counterclockwise). the plane. For each point x in the plane, we write α(x) = xOA α(x) Let C(x) be the circle of center O and radius |Ox| + |Ox| . If the points in the plane are colored with finitely many colors, then there exists a point y with α(y) > 0 such that the color of y also belongs to the circle C(y). Problem 2.63. (Amer. Math. Monthly) Let k, n ∈ Z+ . 1. Assume that n − 1 ≤ k ≤ n(n−1) 2 . Show that there exist n distinct points x1 , ..., xn on a line that determine exactly k distinct distances |xi − xj |. 2. Suppose that [ n2 ] ≤ k ≤ n(n−1) 2 . Then there exist n points in the plane that determine exactly k distinct distances. 3. Prove that for any ε > 0, there exists some constant n0 = n0 (ε) such that for any n > n0 and εn < k < n(n−1) 2 , there exist n points in the plane that determine exactly k distinct distances. Problem 2.64. (International Math. Olympiad 1978) Show that it is possible to pack 2n(2n + 1) nonoverlapping pieces having the form of a parallelepiped of dimensions 1 × 2 × (n + 1) in a cubic box of side 2n + 1 if and only if n is even or n = 1.

20

2 Algebra and Combinatorics

Problem 2.65. (Putnam Competition 1964) Let F be a finite subset of R with the property that any value of the distance between two points from F (except for the largest one) is attained at least twice, i.e., for two distinct pairs of points. Prove that the ratio of any two distances between points of F is a rational number.

3 Geometry

3.1 Synthetic Geometry Problem 3.1. Let I be the center of the circle inscribed in the triangle ABC and consider the points α, β, γ situated on the perpendiculars from I on the sides of the triangle ABC such that |I α| = |Iβ| = |I γ |. Prove that the lines Aα, Bβ, Cγ are concurrent. Problem 3.2. We consider the angle xOy and a point A ∈ Ox. Let (C) be an arbitrary circle that is tangent to Ox and Oy at the points H and D, respectively. Set AE for the tangent line drawn from A to the circle (C) that is different from AH . Show that the line DE passes through a fixed point that is independent of the circle (C) chosen above. Problem 3.3. Let C be a circle of center O and A a fixed point in the plane. For any with point P ∈ C, let M denote the intersection of the bisector of the angle AOP the circle circumscribed about the triangle AOP . Find the geometric locus of M as P runs over the circle C. Problem 3.4. Let ABC be an isosceles triangle having |AB| = |AC|. If AS is an interior Cevian that intersects the circle circumscribed about ABC at S, then describe the geometric locus of the center of the circle circumscribed about the triangle BST , where {T } = AS ∩ BC. Problem 3.5. Let AB, CD, EF be three chords of length one on the unit circle. Then the midpoints of the segments |BC|, |DE|, and |AF | form an equilateral triangle. Problem 3.6. (Amer. Math. Monthly) Denote by P the set of points of the plane. Let : P × P → P be the following binary operation: A B = C, where C is the unique point in the plane such that ABC is an oriented equilateral triangle whose orientation is counterclockwise. Show that is a nonassociative and noncommutative operation satisfying the following “medial property”:

22

3 Geometry

(A B) (C D) = (A C) (B D). Problem 3.7. Consider two distinct circles C1 and C2 with nonempty intersection and let A be a point of intersection. Let P , R ∈ C1 and Q, S ∈ C2 be such that P Q and RS are the two common tangents. Let U and V denote the midpoints of the chords P R and QS. Prove that the triangle AU V is isosceles. Problem 3.8. (Amer. Math. Monthly) If the planar triangles AU V , V BU , and U V C are directly similar to a given triangle, then so is ABC. Recall that two triangles are directly similar if one can obtain one from the other using a homothety with positive ratio, rotations and translations. Problem 3.9. (Amer. Math. Monthly) Find, using a straightedge and a compass, the directrix and the focus of a parabola. Recall that the parabola is the geometric locus of those points P in the plane that are at equal distance from a point O (called the focus) and a line d called the directrix. Problem 3.10. Prove that if M is a point in the interior of a circle and AB ⊥ CD are two chords perpendicular at M, then it is possible to construct an inscribable quadrilateral with the following lengths: |AM| − |MB|, |AM| + |MB|, |DM| − |MC|, |DM| + |MC|. Problem 3.11. (Amer. Math. Monthly) If the Euler line of a triangle passes through the Fermat point, then the triangle is isosceles. Problem 3.12. Consider a point M in the interior of the triangle ABC, and choose A ∈ AM, B ∈ BM, and C ∈ CM. Let P , Q, R, S, T , and U be the intersections of the sides of ABC and A B C . Show that P S, T Q, and RU meet at M. Problem 3.13. Show that if an altitude in a tetrahedron crosses two other altitudes, then all four altitudes are concurrent. Problem 3.14. Three concurrent Cevians in the interior of the triangle ABC meet the corresponding opposite sides at A1 , B1 , C1 . Show that their common intersection point is uniquely determined if |BA1 |, |CB1 |, and |AC1 | are equal. Problem 3.15. (International Math. Olympiad 1983) Let ABCD be a convex quadrilateral with the property that the circle of diameter AB is tangent to the line CD. Prove that the circle of diameter CD is tangent to the line AB if and only if AD is parallel to BC. Problem 3.16. Let A , B , C be points on the sides BC, CA, AB of the triangle ABC. Let M1 , M2 be the intersections of the circle A B C with the circle ABA and let N1 , N2 be the analogous intersections of the circle A B C with the circle ABB . 1. Prove that M1 M2 , N1 N2 , AB are either parallel or concurrent, in a point that we denote by A1 ;

3.2 Combinatorial Geometry

23

2. Prove that the analogously defined points A1 , B1 , C1 are collinear. Problem 3.17. (Putnam Competition) A circumscribable quadrilateral of area S = √ abcd is inscribable. Problem 3.18. (Nieuw Archief v. Wiskunde) Let O be the center of the circumcircle, Ge the Gergonne point, N a the Nagel point, and G1 , N1 , the isogonal conjugates of G and N, respectively. Prove that G1 , N1 , and O are collinear (see also the Glossary for definitions of the important points in a triangle).

3.2 Combinatorial Geometry Problem 3.19. Consider a rectangular sheet of paper. Prove that given any ε > 0, one can use finitely many foldings of the paper along its sides in either 2 equal parts or 3 equal parts to obtain a rectangle whose sides are in ratio r for some r satisfying 1 − ≤ r ≤ 1 + . Problem 3.20. (Amer. Math. Monthly) Show that there exist at most√three points on the unit circle with the distance between any two being greater than 2. Problem 3.21. (Komal) A convex polygon with 2n sides has at least n diagonals not parallel to any of its sides. Problem 3.22. Let d be the sum of the lengths of the diagonals of a convex polygon P1 . . . Pn and p its perimeter. Prove that for n ≥ 4, we have n n + 1 d n−3 10−21 , because 1/|Fi | > 10−7 . Problem 2.16. Prove that if n > 2, then we do not have any nontrivial solutions for the equation x n + y n = zn , where x, y, z are rational functions. Solutions of the form x = af, y = bf, z = cf , where f is a rational function and a, b, c are complex numbers satisfying a n +bn = cn , are called trivial.

6.1 Algebra

93

Solution 2.16. Any rational solution yields a polynomial solution, by clearing the denominators. Let further (f, g, h) be a polynomial solution for which the degree r = max{deg(f ), deg(g), deg(h)} is minimal among all polynomial solutions, and where r > 0. Consider the root of unity ξ = exp 2πi n and assume that f, g, and h are relatively prime. The equation can be written as n−1 &

(f − ξ j h) = g n .

j =0

Now, gcd(f, h) and gcd(f, g) are relatively prime, so that f − ξ j h and f − ξ k h are relatively prime when k = j . Since the factorization of polynomials is unique, we must have g = g1 · · · gn−1 , where gj are polynomials satisfying gjn = f − ξ j h. Consider now the set {f − h, f − ξ h, f − ξ 2 h}. Since n > 2, these elements belong to the two-dimensional space generated by f and h over C. Thus there exists a vanishing linear combination with complex coefficients in these three elements. √ Thus, there exist ai ∈ C so that a0 g0n + a1 g1n = a2 g2n . We then set hj = n aj gj , and observe that hn0 + hn1 = hn2 . Moreover, the polynomials hi and hj are relatively prime if i = j and max deg(hi ) < r, which contradicts our choice of r. This proves the claim. Problem 2.17. A table is an n × k rectangular grid drawn on the torus, every box being assigned an element from Z/2Z. We define a transformation acting on tables as follows. We replace all elements of the grid simultaneously, each element being changed into the sum of the numbers previously assigned to its neighboring boxes. Prove that iterating this transformation sufficiently many times, we always obtain the trivial table filled with zeros, no matter what the initial table was, if and only if n = 2p and k = 2q , for some integers p, q. In this case, we say that the respective n × k grid is nilpotent. Solution 2.17. Consider the m × m square matrices (m ≥ 2) ⎛ ⎞ 0 1 0 0 ··· 0 1 ⎜1 0 1 0 ··· 0 0⎟ ⎜ ⎟ ⎜0 1 0 1 ··· 0 0⎟ ⎜ ⎟ Dm = ⎜ . . . . . . . ⎟ . ⎜ .. .. .. .. .. .. .. ⎟ ⎜ ⎟ ⎝0 0 0 0 ··· 0 1⎠ 1 0 0 0 ··· 1 0 We will write ≡ below if the equality holds (only) for matrices of entries modulo 2. Then the transformation T from the statement acts on the vector space Mn,k (Z/2Z) of n × k matrices with entries in Z/2Z as follows: T (X) = Dn X + XDk , for X ∈ Mn,k (Z/2Z).

94

6 Algebra and Combinatorics Solutions

By induction on s, we have that s

s

s

T 2 (X) ≡ Dn2 X + XDk2 . Thus there exists N such that T N (X) ≡ 0 if and only if there exists some s such that s s Dn2 X ≡ XDk2 . Moreover, we can prove easily that for two matrices A and B of appropriate sizes, we have AX ≡ XB for all X ∈ Mn,k (Z/2Z) if and only if A and B are multiples of the identity by the same scalar. This follows, for instance, if we take X having all but one column (or line) trivial. This assertion is a special case of Schur’s lemma. s Thus T is nilpotent if and only if there exist s and β ∈ Z/2Z such that Dn2 = β ·1n s and Dk2 = β · 1k . Thus the n × k grid is nilpotent if and only if the n × 1 and 1 × k grids are nilpotent. Let us analyze the case of the n × 1 grid. Set a = (a1 , . . . , an ) and a[k] = Dnk a = [k] a1 , . . . , an[k] . By induction, we have k

as[2 ] ≡ as+2k−1 + as−2k−1 , q

where the indices are modulo n. In particular, if n = 2q , then Dn2 = 0 and the grid is nilpotent. This proves the “if” part of the statement. Assume next that n is odd and there exists some r such that a[r] ≡ 0, for any a. Take such an r that is minimal. If n ≥ 3, then r ≥ 3. We claim now that as[r−1] ≡ 1 for [r−1] [r−1] all s. This is true at least for one s, since r is minimal. Since as[r] ≡ as+1 + as−1 , we find that the terms corresponding to even indices (modulo n) are all equal. But n [r−2] [r−2] + as+1 . We is odd and thus all terms are equal. Further, we have as[r−1] = as−1 obtain then that n≡

n

as[r−1] =

s=1

n

n [r−2] [r−2] + as+1 as[r−2] ≡ 0 (mod 2), as−1 =2

s=1

s=1

which is a contradiction. This proves actually that a[r] ≡ 0 when n is odd if and only if a ∈ {(0, 0, . . . , 0), (1, 1, . . . , 1)}. Finally, consider n = 2m h, with h odd. For a vector a of length n, we set s (a) = (as , as+2m , . . . , as+(h−1)2m ). We observe now that

m

s (T 2 (a)) = T (s (a)) m+1

because as[2 ] = as + as+2m . Since the vector s (a) is of length h (which is odd), the previous claim shows us that T N (s (a)) ≡ 0 for some N only if s (a) ∈ {(0, 0, . . . , 0), (1, 1, . . . , 1)}. Moreover, this happens for all values of s. This condition is trivial only when h = 1. This proves the “only if” part. Comments 37 This result was stated in the 1980s as an open question in the journal Kvant. The same method can be used to show that the n × k grids in the plane are

6.2 Algebraic Combinatorics

95

nilpotent if and only if n = 2p − 1 and k = 2q − 1 for some natural numbers p, q. A nice corollary is the following. Define the sequence (αk ) by the recurrence relations α1 = 0, αk+1 = min(2αk + 2, 2N − 2αk − 1). Then there exists some k such that αk = N − 1 if and only if the parameter N has the form 2q − 1 for some natural number q. The solution given here is from: •

V. Boju, L. Funar: Iterative processes for Zn2 , Analele Univ. Craiova 15 (1987), 33–38.

6.2 Algebraic Combinatorics Problem 2.18. Let us consider a four-digit number N whose digits are not all equal. We first arrange its digits in increasing order, then in decreasing order, and finally, we subtract the two obtained numbers. Let T (N) denote the positive difference thus obtained. Show that after finitely many iterations of the transformation T , we obtain 6174. Solution 2.18. Let N have the digits a, b, c, d. 1. Assume that a ≥ b = c ≥ d. Then in T (N), the sum of the first and the fourth digits is 9, as well as the sum of the second and the third. These give five combinations of four digits that give 6174, by applying T once more. 2. Otherwise, a ≥ b > c ≥ d. Then in T (N), the sum of the digits placed at extremities is 10, while the sum of the middle ones is 8. This gives 25 combinations all leading to the desired result, namely 6174. Comments 38 Let a be a positive integer having r digits, not all equal, in base g. Let a be the g-adic integer formed by arranging the digits of a in descending order and let a be that formed by ascending order. Define T (a) = a − a . Then a is said to be self-producing if T (a) = a. There exists an algorithm for obtaining self-producing numbers in any base. Such a fixed point k of T is called a (g-adic) Kaprekar constant if it has the further property that every r-digit integer a eventually yields k on repeated iteration of T and moreover, k is fixed by T . D.R. Kaprekar found that 6174 has this nice property in a short note published in 1949. Ludington proved that for large r ≥ ng there does not exist a Kaprekar constant. Here ng is given by ⎧ if g = 2k, ⎨ g − 1 + (k − 1)δ, ng = 2k + 1 + (g − k − 2)δ, if g = 2t k + 2t−1 + 1, ⎩ 3k − 4 + (g − 1)δ, if g = 2t + 1, and δ ∈ {0, 1} is equal to g (mod 2). Further improvements by Lapenta, Ludington, and Prichett showed that there is no Kaprekar constant in base 10 when the number of digits is r > 4.

96

6 Algebra and Combinatorics Solutions

Hasse and Prichett found that there exists a 4-digit Kaprekar constant in base g if and only if g = 2n · 5, where n is either 0 or an odd number. Moreover, the situation is completely understood for If the base g 3-digit numbers. r is odd, then there is a Kaprekar constant given by r−2 , r − 1, , which is reached 2 2 r+2 within 2 iterations of T (respectively 1, if r = 2). For instance, 495 is a Kaprekar constant if g = 10. If the base g is even, then there is no Kaprekar constant. More precisely, iterations of T willeventually reach the 2 formed by the pair r−1 loop of length r+1 r−1 r+1 of numbers r−3 , , after at most , r − 1, , r − 1, 2 2 2 2 2 steps (respectively 1 if r = 3). •

• • • •

K.E. Eldridge and S. Sagong: The determination of Kaprekar convergence and loop convergence of all three-digit numbers, Amer. Math. Monthly 95 (1988), 105–112. J.F. Lapenta, A.L. Ludington, and G.D. Prichett: An algorithm to determine selfproducing r-digit g-adic integers, J. Reine Angew. Math. 310 (1979), 100–110. G.D. Prichett, A.L. Ludington, and J.F. Lapenta: The determination of all decadic Kaprekar constants, Fibonacci Quart. 19 (1981), 45–52. A.L. Ludington: A bound on Kaprekar constants, J. Reine Angew. Math. 310 (1979), 196–203. H. Hasse and G.D. Prichett: The determination of all four-digit Kaprekar constants, J. Reine Angew. Math. 299/300 (1978), 113–124.

Problem 2.19. Find an example of a sequence of natural numbers 1 ≤ a1 < a2 < · · · < an < an+1 < · · · with the property that every m ∈ Z+ can be uniquely written as m = ai − aj , for i, j ∈ Z+ . Solution 2.19. We consider the sequence a1 = 1, a2 = 2, a2n+1 = 2a2n , a2n+2 = a2n+1 + rn , where rn is the smallest natural number that cannot be written in the form ai − aj , with i, j ≤ 2n + 1. Comments 39 One does not know the minimal growth of such a sequence ak . Problem 2.20. Consider the set of 2n integers {±a1 , ±a2 , . . . , ±an } and m < 2n . Show that we can choose a subset S such that 1. The two numbers ±ai are not both in S; 2. The sum of all elements of S is divisible by m. Solution 2.20. Let S1 , . . . , S2n −1 be the 2n − 1 nonempty distinct subsets of the set {a1 , . . . , an }, where ai ≥ 0. Let F (Si ) denote the sum of the elements of Si . Then, by the pigeonhole principle, there exist i, j such that F (Si ) ≡ F (Sj ) (mod m). Consider next the set S = {Si \ Sj } ∪ −{Sj \ Si }. We derive that F (S) ≡ 0 (mod m).

6.2 Algebraic Combinatorics

97

Problem 2.21. Show that for every natural number n there exist prime numbers p and q such that n divides their difference. Solution 2.21. Consider the following arithmetic progression: 1, 1+n, 1+2n, . . . , 1+ rn, . . . . According to Dirichlet’s theorem, there exist infinitely many prime numbers among the terms of this progression. Let p, q be two of these prime numbers; then p = 1 + nr and q = 1 + ns, where r = s. This yields n(r − s) = p − q, as claimed. An alternative solution is to consider the set of n arithmetic progressions of ratio n starting respectively at 0, 1, 2, . . . , n − 1. Since the set of prime numbers is infinite (this is elementary), there exists at least one progression having infinitely many prime numbers among its terms. The argument above settles the claim. Problem 2.22. An even number, 2n, of knights arrive at King Arthur’s court, each one of them having at most n − 1 enemies. Prove that Merlin the wizard can assign places for them at a round table in such a way that every knight is sitting only next to friends. Solution 2.22. 1. We consider the friendship graph G defined below: its vertices are in bijection with the knights and the edges are joining pairs of vertices whose respective knights are not enemies. The degree of each vertex is at least n. According to Dirac’s theorem, such a graph admits a Hamiltonian cycle, and this yields the wanted assignment of places. 2. Choose an arbitrary assignment of places in which we have two neighbors, A and B, who are enemies. Let us assume that A lies on the right-hand side of B. ˜ B, ˜ According to the pigeonhole principle, there exists another pair of enemies, say A, ˜ who are neighbors, and moreover, A˜ is on the right of B. Let us switch all the places starting at A (and lying on the right side of A) and ˜ using a symmetry. After such a transformation, the number of enemy ending at B, pairs (A, B) is diminished by 2. Applying such transforms iteratively, one obtains a position in which all neighbors are friends. Problem 2.23. Let r, s ∈ Z+ . Find the number of 4-tuples of positive integers (a, b, c, d) that satisfy 3r 7s = lcm(a, b, c) = lcm(a, b, d) = lcm(a, c, d) = lcm(b, c, d). Solution 2.23. The numbers a, b, c, d are of the form 3mi 7ni , 1 ≤ i ≤ 4, where 0 ≤ mi ≤ r and 0 ≤ ni ≤ s. Also, mi = r for at least two values of i, and ni = s for at least two values of i. We have then: (1) one possibility that mi = r for all four i; (2) 4r possibilities that precisely one mi ∈ {0, . . . , r − 1}; that exactly (3) C42 r 2 possibilities two mi ∈ {0, . . . , r − 1}. Therefore, there are 1 + 4r + 6r 2 possibilities for the mi ’s and a similar number for the ni ’s, yielding a total of 1 + 4r + 6r 2 1 + 4s + 6s 2 possibilities. Problem 2.24. 1. Let n ∈ Z+ and p be a prime number. Denote by N (n, p) the number of binomial coefficients Cns that are not divisible by p. Assume that n is written in base p as n = n0 + n1 p + · · · + nm p m , where 0 ≤ nj < p, for all j ∈ {0, 1, . . . , m}. Prove that N (n, p) = (n0 + 1)(n1 + 1) · · · (nm + 1).

98

6 Algebra and Combinatorics Solutions

2. Write k in base p as k = k0 + k1 p + · · · + ks p s , with 0 ≤ kj ≤ p − 1, for all j ∈ {0, 1, . . . , s}. Prove that Cnk ≡ Cnk00 Cnk11 · · · Cnkss (mod p). Solution 2.24. 1. It is clear that Cpk ≡ 0 (mod p) for all k ∈ {1, 2, . . . , p}. Thus n n (1 + x)p ≡ 1 + x p in (Z/pZ)[x]. By induction, we find that (1 + x)p ≡ 1 + x p in (Z/pZ)[x] for any natural number n. We have then the following congruences in (Z/pZ)[x]: m

m

(1+x)n = (1+x)n0 (1+x)n1 p · · · (1+x)nm p ≡ (1+x)n0 (1+x p )n1 · · · (1+x p )nm . When developing factors on the right-hand side, we obtain a sum of factors 2 m x a0 +a1 p+a2 p +···+am p , with a coefficient that is nonzero modulo p. Since every number can be uniquely written in base p, all these factors are distinct and their coefficients modulo p are nonzero. There are exactly (n0 + 1)(n1 + 1) · · · (nm + 1) such factors, and therefore as many binomials not divisible by p. 2 m 2. As observed above, the factor x k appears in the form x a0 +a1 p+a2 p +···+am p ; since k can be uniquely written in base p, we have ai = ki . This implies that the k coefficient of x k is Cnk00 Cnk11 · · · Cnjj , and hence the claim. Problem 2.25. Define the sequence Tn by T1 = 2, Tn+1 = Tn2 − Tn + 1, for n ≥ 1. Prove that if m = n, then Tm and Tn are relatively prime, and further, that ∞ 1 = 1. Ti i=1

Solution 2.25. We prove by induction that Tn+1 = 1 + T1 T2 · · · Tn . In fact, we have Tn+1 = Tn2 − Tn + 1 = Tn (Tn − 1) + 1 = Tn (Tn−1 Tn−2 · · · T1 ) + 1 = 1 + T1 · · · Tn . Then take m < n. Therefore, Tm divides T1 · · · Tn−1 = Tn −1 and thus gcd(Tm , Tn ) = 1. Further, we prove by induction that n 1 1 =1− . Ti Tn+1 − 1 i=1

In fact, k+1 1 1 1 1 1 =1− =1− + =1− . Ti Tk+1 − 1 Tk+1 Tk+1 (Tk+1 − 1) Tk+2 − 1 i=1

Since Tn tends to infinity, we therefore obtain

∞

1 i=1 Ti

= 1.

6.2 Algebraic Combinatorics

99

Problem 2.26. Let α, β > 0 and consider the sequences [α], [2α], . . . , [kα], . . . ; [β], [2β], . . . , [kβ], . . . , where the brackets denote the integer part. Prove that these two sequences taken together enumerate Z+ in an injective manner if and only if α, β ∈ R \ Q and

1 1 + = 1. α β

Solution 2.26. Set AN = {1, 2, . . . , N}; take k maximal such that kα < N + 1 and l maximal such that the following inequalities hold: 1. Therefore, lβ < N + N N+1 N N+1 ≤ l ≤ . ≤ k ≤ and α α β β Since AN is injectively enumerated by the two sequences, we have k + l = N , and hence

N N N +1 N +1 + ≤N ≤ + . α β α β Letting N go to infinity, we obtain α1 + β1 = 1. p If α ∈ Q, then also β ∈ Q. In this case, write α = m n and β = q . It follows that [αnqp] = [βmqn], which contradicts the injectivity assumption. Therefore α ∈ Q and β ∈ Q. Conversely, we have

N +1 N +1 N +1 N +1 N +1 N +1 + > + > + −2 = N −1, N +1 = α β α β α β

whence

N +1 N +1 + = N. α β

In particular, using the notation from above, we have k + l = N . If the two sequences enumerate AN in a surjective manner, then they will also enumerate AN injectively, because k +l = N . It suffices then to prove the surjectivity. Let us assume the contrary. Then there exist u, x, y ∈ Z+ such that xα < u < u + 1 < (x + 1)α and yβ < u < u + 1 < (y + 1)β. Dividing by α, β respectively and adding up the inequalities, we obtain x+y

1 and α1 , . . . , αm ∈ R+ . Then the sets

100

6 Algebra and Combinatorics Solutions

Si = {[nαi ], where n ∈ Z+ } form a complementary system only if m = 2, α1−1 + α2−1 = 1, and α1 ∈ R \ Q. Solution 2.27. If all αj < 1, then the collection Sj contains twice some number. Let 1 ∈ S1 , and thus [α1 ] = 1, α1 = 1. We prove first that S1 = {[nα1 ]; n ∈ Z+ } and T = {[nβ − ]; n ∈ Z+ } form a complementary system, where ' α1 (2(a − c))−1 , if α1 = ac ∈ Q, β= , and = 0, otherwise. α1 − 1 Let = {nα1 , nβ − ; n ∈ Z+ }. It is sufficient to show that for any integer M > 1, we have the formula N = card({x ∈ ; x < M}) = M − 1. Now, nα1 < M is equivalent to nα1 + δ < M, where δ = (2c)−1 if α1 = a/c ∈ Q, while δ = 0, if α ∈ R \ Q. The maximum value for n is therefore M−δ α1 . In a similar way, the number of elements of the form nβ − that are less than M is M+ . This β implies that N = [(M − δ)/α1 ] + [(M + )/β]. Now we have

M −δ M −δ M −δ < −1< , α1 α1 α1

M + M + M + −1< < , β β β and by summing up the two inequalities, we obtain M − 2 < N < M, and therefore N = M − 1, as claimed. Now let k be the smallest integer k ∈ S1 . Then there exists some α2 such that k = [α2 ] = [β − ] ≥ 2. We have nβ − 1 ≤ [nβ − ] < nβ − , and therefore [(n + 1)β − ] − [nβ − ] < (n + 1)β − − nβ + 1 = β + 1 − ≤ k + 2 − , [(n + 1)β − ] − [nβ − ] > (n + 1)β − 1 − nβ + = β − 1 + > k − 1 + 2 . Therefore k ≤ [(n + 1)β − ] − [nβ − ] = [(n + 1)α2 ] − [nα2 ] ≤ k + 1. The difference between two consecutive terms of the sequence [nβ − ] is equal to the difference between consecutive missing terms from the sequence [nα1 ]; that is, k or k + 1 is the same as the difference between two consecutive terms of the sequence [nα2 ]. This implies the fact that the j th term that does not belong to S1 is precisely [j α2 ] = [jβ − ]. This implies that m = 2. If α1 ∈ Q, then α2 = b/d, and for j = d, we obtain [j α2 ] > [jβ − ], which is false.

6.2 Algebraic Combinatorics

101

Comments 40 The fact that the two sequences from the previous problem are complementary is known as the Rayleigh–Beatty theorem, since S. Beatty proposed it as a problem in Amer. Math. Monthly in 1926. Presumably, this was known to Rayleigh, who mentioned it without proof in 1894. In 1927, J.V. Uspensky proved that if the m sequences are complementary, then m ≤ 2; his proof was simplified in several papers by Skolem, Graham, and Fraenkel, who also provided a far-reaching generalization in 1969. • • • • • • •

S. Beatty: Problem 3173, Amer. Math. Monthly 33 (1926), 3, 156. J. Lambek, L. Moser: Inverse and complementary sequences of natural numbers, Amer. Math. Monthly 61 (1954), 454–458. A.S. Fraenkel: Complementary systems of integers, Amer. Math. Monthly 84 (1977), 114–115. A.S. Fraenkel: The bracket function and complementary sets of integers, Canad. J. Math. 21 (1969), 6–27. R.L. Graham: On a theorem of Uspensky, Amer. Math. Monthly 70 (1963), 407– 409. Th. Skolem: On certain distributions of integers in pairs with given differences, Math. Scand. 5 (1957), 57–68. J.V. Uspensky, M.A. Heaslet: Elementary Number Theory, McGraw-Hill, New York, 1939.

Problem 2.28. Let f : Z+ → Z+ be an increasing function and set F (n) = f (n) + n, G(n) = f ∗ (n) + n, where f ∗ (n) = card({x ∈ Z+ ; 0 ≤ f (x) < n}). Then {F (n); n ∈ Z+ } and {G(n); n ∈ Z+ } are complementary sequences. Conversely, any two complementary sequences can be obtained this way using some nondecreasing function f . Solution 2.28. We compute the number N of integers from {F (n), G(n), n ∈ Z+ } that are smaller than M. Suppose that k terms of the form G(n) are smaller than M. Then f ∗ (k) + k < M ≤ f ∗ (k + 1) + k + 1, and so f ∗ (k) < M − k < f ∗ (k + 1) + 1. Since f ∗ (k) = card{x ∈ Z+ ; f (x) < k} < M − k, we derive f (M − k) ≥ k. Further, f ∗ (k + 1) = {x ∈ Z+ ; f (x) < k + 1} ≥ M − k − 1 implies that f (M − k − 1) < k + 1. Therefore F (M − k) = f (M − k) + M − k ≥ M and F (M − k − 1) < M. This means that exactly M − k terms of the form F (n) are smaller than M; therefore N = M. The converse is immediate by defining f (n) = F (n) − n. Comments 41 Since f ∗ is nondecreasing, it follows that it makes sense to define f ∗∗ . The sequences G(n) and H (n), where H (n) = f ∗∗ (n) + n, are complementary, and thus f ∗∗ = f . The result is due to A. Frankel.

102

6 Algebra and Combinatorics Solutions

Problem 2.29. Let M denote the set of bijective functions f : Z+ → Z+ . Prove that there is no bijective function between M and Z. Solution 2.29. We will find an injection ρ : R \ (Q ∪ [0, 2]) → M, as follows. To any α ∈ R \ (Q ∪ [0, 2]) there is associated an irrational β < 2 such that α1 + β1 = 1. Let ρ(α) : Z+ → Z+ be the function defined by ' n α , for even n, ρ(α)(n) = 2n+1 β 2 , for odd n. According to the previous problem, ρ(α) : Z+ → Z+ is a bijection. Moreover, the map ρ is easily seen to be injective. Thus card(M) ≥ card(R \ Q) = card(R) > card(Z), and the result follows. Problem 2.30. Let F ⊂ Z be a finite set of integers satisfying the following properties: 1. For any x ∈ F , there exist y, z ∈ F such that x = y + z. 2. There exists n such that, for any natural number 1 ≤ k ≤ n, and any choice of x1 , . . . , xk ∈ F , their sum x1 + · · · + xk is nonzero. Prove that card(F ) ≥ 2n + 2. Solution 2.30. By hypothesis, 0 ∈ F . Let F+ = F ∩ Z+ , F− = F ∩ Z− , so that F = F+ ∪ F− . Consider the unoriented graph whose vertices are the elements of F+ and whose edges are defined below: x and y are adjacent if there exists z ∈ F such that x = y + z. By the first hypothesis, each vertex is adjacent to at least one edge. This implies that the graph contains a cycle [x1 , . . . , xk ]. Assume that k is minimal with this property. This means that there exist zi ∈ F such that: x1 = x2 + z1 = x3 + z2 + z1 = · · · = x1 + zk + · · · + z1 , which implies that z1 + · · · + zk = 0. The second hypothesis implies that k ≥ n + 1, and thus card(F+ ) ≥ k ≥ n + 1. A similar argument shows that card(F− ) ≥ n + 1, and the claim follows. Problem 2.31. For a finite graph G we denote by Z(G) the minimal number of colors needed to color all its vertices such that adjacent vertices have different colors. This is also called the chromatic number of G. Prove that the inequality Z(G) ≥ holds if G has p vertices and q edges.

p2

p2 − 2q

6.2 Algebraic Combinatorics

103

Solution 2.31. If we fix N = χ (G) and the number p of vertices of the graph G, then we have to show that the number q of edges allowed p2 1 q≤ 1− . 2 N By hypothesis, there exists a partition (according to the colors) of the set of vertices into N classes such that two vertices in the same class are not adjacent. Let n1 , n2 , . . . , nN be the number of vertices in the respective classes. The only possibility to get an edge is to join two vertices lying in different classes, and thus the total number q of edges is at most 1≤i<j ≤N ni nj . Further, one knows that 1≤i≤N ni = p. We now have

ni nj ≤

2 CN

i<j

p2 p2 = N2 2

1 1− N

.

The equality holds if and only if N divides p and ni = p/N , for all i. Comments 42 It can be proved, in the same way, that we have p t − [t] χ (G) ≥ 1− , [t] 1 + [t] where t = p −

2q p .

There exists also an upper bound for an arbitrary graph, namely χ (G) ≤

1 1 + 8q + 1 . 2

One problem that received a great deal of attention in the past was to get bounds for the chromatic number of a graph in terms of its topology. For instance, assume that the graph G is drawn on some surface so that its edges are not crossing each other, in which case the graph is said to be embedded. If the surface is partitioned into curvilinear polygons (called “countries”), then the dual graph of this decomposition is constructed by associating a vertex to each country and joining two vertices if the respective countries have a common frontier. This way one constructs all graphs embedded in that surface. Surfaces in R3 are characterized by only one number, called the genus g, or equivalently, by its Euler–Poincaré characteristic, which is 2 − 2g. The latter is computed elementarily starting from an arbitrary partition into curvilinear polygons. If such a partition consists of V vertices, E edges, and F polygonal countries, then 2 − 2g = V − E + F. Computing a bound for the chromatic number of an arbitrary graph embedded in a surface is equivalent to finding the number of colors needed for coloring any map of the surface so that adjacent countries have different colors. Now, if G is a graph that can be embedded in a surface of genus g, then it was stated without proof by Heawood in 1890 that

104

6 Algebra and Combinatorics Solutions

χ (G) ≤

1 (7 + 1 + 48g) . 2

This has been known as the Heawood conjecture since then, and it was finally proved in 1968 for all surfaces except for the plane by Ringel and Youngs. The case of the plane remained open until 1977, when Appel and Haken presented a computer-assisted proof of that case, thenceforth known as the “four color theorem.” The article of Stahl presents the history of the problem and of the proofs. •

• •

K. Appel, W. Haken: Every Planar Map is Four Colorable, With the collaboration of J. Koch, Contemporary Mathematics, 98, American Mathematical Society, Providence, RI, 1989. G. Ringel: Map Color Theorem, Die Grundlehren der mathematischen Wissenschaften, Band 209, Springer-Verlag, 1974. S. Stahl: The other map coloring theorem, Math. Magazine 58 (1985), 3, 131–145.

Problem 2.32. Let Dk be a collection of subsets of the set {1, . . . , n} with the property that whenever A = B ∈ Dk , then card(A ∩ B) ≤ k, where 0 ≤ k ≤ n − 1. Prove that card(Dk ) ≤ Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . Solution 2.32. Let Pk be the set of those subsets of {1, . . . , n} with at most k elements. Then any maximal (with respect to inclusion) collection Dk as above must contain Pk , since otherwise, Dk ∪ Pk will still satisfy the requirements while being strictly *k = Dk \ Pk . For larger than Dk . Therefore, let us consider a maximal Dk and set D + = {A1 , . . . , As | card(Ai ) = k + 1, and Ai ⊂ A}. If B ∈ D *k *k , let us set D A∈D +∩ B + = ∅, then there exist i, j such that Ai = Bj , where card(Ai ) = k + 1; and A this would imply that A ∩ B ⊃ Ai and thus card(A ∩ B) ≥ k + 1, which is absurd. + for A ∈ D *k , are disjoint. Since each A + has at least one Therefore, the sets A, *k ) ≤ card (Pk+1 \ Pk ) and hence + ⊂ Pk+1 \ Pk , we obtain that card(D element and A *k ) + card(Pk ) = card(Pk+1 ) = Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . We card(Dk ) ≤ card(D have equality for Dk = Pk+1 . Problem 2.33. Prove that

⎧ 0, ⎪ ⎪ ⎪ ⎪ 1, ⎪ ⎪ ⎪ ⎨ n/2,

if if n if 1 (−1)n−k Cnk k p = n(3n+1) , if ⎪ p! 24 ⎪ k=0 ⎪ n2 (n+1) ⎪ ⎪ if ⎪ 48 , ⎪ ⎩ n(15n3 +30n2 +5n+1) , if 1152

0 ≤ p < n, p = n, p = n + 1, p = n + 2, p = n + 3, p = n + 4.

Solution 2.33. Consider the function (ex −1)n , in which we develop first the binomial and then we develop the exponentials in Taylor series:

6.2 Algebraic Combinatorics

(ex − 1)n =

n

(−1)n−k Cnk ekx =

k=0

n

(−1)n−k Cnk

∞ j =0

k=0

n ∞ 1 n−k k j (−1) Cn k . = j! j =0

1 j j k x j!

105

k=0

For the same function we now develop the exponential and then the binomial as follows: n 1 1 (ex − 1)n = x + x 2 + x 3 + · · · 2! 3! n n+1 n(3n + 1) n+2 n2 (n + 1) n+3 x x x + + 2 24 48 n(15n3 + 30n2 + 5n + 1) n+4 + + ··· . x 1152

= xn +

By identifying the first coefficients in the two series, we obtain the formulas from the statement. Comments 43 The problem of computing the sums sn,p = nk=0 (−1)k Cnk k p is a classical one, and the first result (for p = n) is due to Leonhard Euler. D. Andrica gave a method of computation by means of the recurrence relations sn,p+1 = n(sn,p − sn−1,p ),

p ≥ 0.

In particular, one recovers the results from the problem. •

D. Andrica: On a combinatorial sum, Gazeta Mat. 5 (1989), 158.

Problem 2.34. Write lcm(a1 , . . . , an ) in terms of the various gcd(ai , . . . , aj ) for subsets of {a1 , . . . , an }. Solution 2.34. Let Pj = 1≤i1 · · · > σ (n). Let us prove first the following intermediate result. If a < c, b < d, then the inequality |a − b| + |c − d| ≤ |a − d| + |b − c| holds with equality when either a < c < b < d or b < d < a < c. From the symmetry of the previous inequality, we can assume that a < b, d, and using a translation followed by a homothety on the real axis, we can, moreover, assume that a = 0, c = 1. If b < 1, d < 1, the inequality reduces to 2b < 2d. If b < 1, d > 1, then the inequality is 2b < 2. If b > 1, d > 1, we obtain equality. This proves the claim.

6.2 Algebraic Combinatorics

107

Now let σ be a permutation such that S(σ ) is maximal. Suppose that there exists i < j such that σ (i) < σ (j ). Let us define another permutation σ ∗ = σ ◦ (i, j ), where (i, j ) denotes the transposition interchanging i and j . According to our claim, we have S(σ ∗ ) ≥ S(σ ). By modifying iteratively σ by taking the product with the transpositions determined by all pairs (i, j ) as above, we will finally obtain γ : 1 2 ··· n γ : . n n − 1 ··· 1 Moreover, the inequality above shows that S(σ ) ≤ S(γ ) =

n

|n − 2i + 1|.

i=1

Problem 2.37. On the set Sn of permutations of {1, . . . , n} we define an invariant distance function by means of the formula d(σ, τ ) =

n

|σ (i) − τ (i)|.

i=1

What are the values that d could possibly take? Solution 2.37. We have d(ρσ, ρτ ) = d(σ, τ ), for any ρ, σ, τ ∈ Sn . Therefore, it suffices to consider the values of d(1, σ ), where 1 is the identity permutation. If m > 0 is a value of d, we will show below that m − 2 is also a value of d, and hence d takes all even values from 0 to some 2t, where t has to be determined. Notice first that d takes only even values, because d(σ, τ ) ≡

n i=1

(σ (i) − τ (i)) ≡

n i=1

σ (i) −

n

τ (i) ≡ 0 (mod 2).

i=1

Let m = d(1, σ ) > 0. There then exist 1 ≤ r < s ≤ n such that σ (r) > r, σ (s) < s, and σ (i) = i, if r < i < s. Let ρr,s be the cycle (r, r + 1, . . . , s). We claim that d(1, ρr,s σ ) = m − 2. First, by hypothesis, σ (r) > s and σ (s) < r. Looking at the contribution of the elements from r to s to the respective distances, we obtain d(1, σ )−d(1, ρr,s σ ) = (σ (r)−r +s−σ (s))−(r −σ (s)+s−r −1+σ (r)−r −1) = 2. 2 Let us show now that t = n4 (see also the previous problem) and the maximum distance d(1, σ ) is realized θ (i) = n + 1 − i. 2for the permutation 2 We have d(1, θ) = 2 n4 = n2 . Let k < n − 1, such that σ (i) = n + 1 − i, for all i < k, while σ (k) = n+1−k. We claim that there exists some permutation σ˜ ∈ Sn with the property that d(1, σ ) ≤ d(1, σ˜ ) and that satisfies σ˜ (i) = n + 1 − i, i ≤ k. Then, using induction k, we find a sequence of permutations reaching θ such that

108

6 Algebra and Combinatorics Solutions

d(1, σ ) ≤ d(1, σ˜ ) ≤ · · · ≤ d(1, θ). If σ (k) = n + 1 − k, we have σ (k) < n + 1 − k because every number greater than n + 1 − k is the image of some i < k. But this implies that σ (r) = n + 1 − k, for some r > k. Let τ be the transposition (k, r). Set σ˜ = τ σ ; we have d(1, σ˜ ) − d(1, σ ) = |r − kσ | + |m + 1 − 2k| − |k − kσ | + |n + 1 − k − r| = w. Let us suppose that k ≤ σ (k) ≤ r ≤ n + 1 − k (the other cases are similar). Then w = r − σ (k) + n + 1 − 2k − σ (k) + k − n − 1 + k + r = 2(r − kσ ) ≥ 0. Analogously, for all k, r we have d(1, σ˜ ) − d(1, σ ) ≥ 0, which proves the claim. Problem 2.38. The set M = {1, 2, . . . , 2n} is partitioned into k sets M1 , . . . , Mk , where n ≥ k 3 + k. Show that there exist i, j ∈ {1, . . . , k} for which we can find k + 1 distinct even numbers 2r1 , . . . , 2rk+1 ∈ Mi with the property that 2r1 − 1, . . . , 2rk+1 − 1 ∈ Mj . 2 Solution 2.38. There exists a set Ms that contains at least 2n k ≥ 2(k + 1) elements. We have to consider two cases: 2 1. Either Ms contains at least 2(k 2+1) = k 2 + 1 even numbers. Then the set of odd numbers O = {r − 1, where r is even, and r ∈ Ms }

has k 2 + 1 elements. Then there exists some set Ma containing at least k k+1 elements from O. We choose therefore i = s and j = a. Notice that i might be equal to j . 2. Or else Ms contains at least k 2 + 1 odd numbers. The solution is similar to that from above, considering the set of even numbers. 2

Problem 2.39. Let S be the set of odd integers not divisible by 5 and smaller than 30m, where m ∈ Z∗+ . Find the smallest k such that every subset A ⊂ S of k elements contains two distinct integers, one of which divides the other. Solution 2.39. Consider the subset N = {1, 7, 11, 13, 17, 19, 23, 29, . . . , 30m − 1} of elements of S that are not divisible by 3. There are 8m elements in N , which are written in increasing order a1 < a2 < · · · < a8m . Every element of S can be uniquely written as x = ai · 3t , where t ∈ Z+ and ai ∈ N . If k ≥ 8m + 1, then according to the pigeonhole principle, any subset A ⊂ S of cardinality card A = k contains two distinct elements x, y ∈ A for which x = ai 3t and y = ai 3q . In this case, either x divides y, or y divides x. Next, for all i, choose the maximal t (i) with the property that ai 3t (i) < 30m < t ai 3 (i)+1 , and set bi = ai 3t (i) . We have therefore 10m < bi < 30m. The set {b1 , b2 , . . . , b8m } contains 8m elements from S, and also we have the inequalities 0 < bi /bj < 3, for any i, j . Since all numbers bi are odd, we derive that bi /bj ∈ Z. Therefore {b1 , . . . , b8m } does not contain a number and one divisor of it. This shows that the required value of k is 8m + 1.

6.2 Algebraic Combinatorics

Problem 2.40. Prove that · · · ≤ an are integers.

ai −aj 1≤j j (ai − aj ) S= = . i−j (n − 1)!(n − 2)! · · · 1! i>j

We now consider the determinant D of the matrix ⎛ 1 1 ··· 1 ⎜ Ca1 Ca1 · · · Ca1 n 1 2 ⎜ ⎜ .. .. .. ⎝ . . .

⎞ ⎟ ⎟ ⎟. ⎠

Can−1 Can−1 · · · Can−1 n 1 2 k using factorials, It is obvious that D ∈ Z+ . We now write the binomial coefficient Cm and we arrange the factors in the lines. We obtain ⎞ ⎛ 1 1 ··· 1 ⎜ a1 a2 · · · an ⎟ ai − aj 1 i>j ⎟ ⎜ det ⎜ .. = S, D= .. .. ⎟ = ⎠ ⎝ . 1!2! · · · (n − 1)! . . i=1 (i − 1)! a1n−1 a2n−1 · · · ann−1

and the claim follows. Problem 2.41. Is there an infinite set A ⊂ Z+ such that for all x, y ∈ A neither x nor x + y is a perfect power, i.e., a k , for k ≥ 2? More generally, is there an infinite set A ⊂ Z+ such that for any nonempty finite collection xi ∈ A, i ∈ J , the sum i∈J xi is not a perfect power? Solution 2.41. 1. Yes. Let p1 = 2, p2 = 3, . . . , pn , . . . be the set of primes in increasing order. We then set A = {22 3, 22 32 5, . . . , 22 32 52 · · · pn2 pn+1 , . . .}. Obviously, no element of A is a perfect power. Moreover, if x ≤ y ∈ A, then we can write x = 22 · · · pk2 pk+1 and x = 22 · · · pn2 pn+1 . If k < n, then pk+1 divides 2 2 · · · pn2 pn+1 , while pk+1 does not divide x + y, x + y = 22 · · · pk2 pk+1 1 + pk+1 pk+2 and thus it cannot be a perfect power. If k = n, the same argument works, unless k = 0, which was excluded from A. 2. Let us prove that if C ⊂ Z+ is a set of density d(C) = 0, then there exists an infinite set A ⊂ Z+ such that for any nonempty finite collection xi ∈ A, i ∈ J , the sum i∈J xi is not in C. Recall that the density of the subset C ⊂ Z+ is defined as d(C) = lim

n→∞

card(C ∩ {1, 2, . . . , n}) . n

Since d(C) = 0, there exists a natural number a1 ∈ C. Then consider C1 = C ∪ (C − a1 ) ∩ Z+ , where we set X − λ = {x − λ|x ∈ X}. The density d has the following fundamental property that results from the definition:

110

6 Algebra and Combinatorics Solutions

d(A ∪ B) ≤ d(A) + d(B). This implies that d(C1 ) = 0, and therefore there exists a natural number a2 ∈ C1 . Assuming that both the subset Ck of density zero and the sequence a1 , a2 , . . . , ak+1 , with ak+1 ∈ Ck , are defined, we set Ck+1 = Ck

k

(Ck − (ai1 + ai2 + · · · + aip )) ∩ Z+ .

p=1 i1 ≤i2 ≤···≤ip ≤k+1

It follows that d(Ck+1 ) = 0 and thus there exists at least one integer ak+2 > ak+1 such that ak+2 ∈ Ck+1 . The sequence (ak ) forms an infinite subset A with the required property. It suffices now to compute the density of the set C = {a k |a ≥ 2, k ≥ 2}. We have √ √ √ [ n] + [ 3 n] + · · · log n n log2 n card(C ∩ {1, 2, . . . , n}) = ≤ = √2 . n n n n Therefore d(C) = 0. Comments 44 There is an analogous result when products are used instead of sums of elements in A. For instance, the set ( ) an a1 a2 A = a1 , a2 = C2a , C , . . . , C a +a a +a n 1 2 n−1 1 has the property that no product of its elements is a perfect power. Is it true that d(A) = 0? Problem 2.42. Let .. A A2 3

f (n) = max A1

Ak

. ,

where n = A1 + · · · + Ak . Thus, f (1) = 1, f (2) = 2, f (3) = 3, f (4) = 4, f (5) = 9, f (6) = 27, f (7) = 512, etc. Determine f (n). A ·· k

Solution 2.42. It is clear that if f (n) = A·1

, then Ai > 1. Also, f is an increasing A ·· k

function, and thus f (n + 1) ≥ f (n) + 1. If A1 = k, then A·2 implies that whenever n ≥ 4, we have f (n) =

≤ f (n − k), and this

max k f (n−k) .

2≤k≤n−2

It is easy now to see that f (n + 1) ≥ 2f (n). In fact, let f (n) = k f (n−k) , for some k ≥ 2; then f (n + 1) ≥ k f (n−k+1) ≥ k f (n−k)+1 ≥ 2f (n). log(a+1) The next step is to compare a f (b+1) with (a + 1)f (b) , that is, f f(b+1) (b) with log(a) . If a ≥ 2, then a + 1 < a 2 and hence log(a + 1)/ log a < 2. Thus, if b ≥ 4, then

6.2 Algebraic Combinatorics

111

f (b + 1)/f (b) ≥ 2, which yields a f (b+1) > (a + 1)f (b) , as soon as a ≥ 2, b ≥ 3. For small values of the arguments, we have log 4 f (4) f (3) log 3 < < < . log 3 f (3) f (2) log 2 These show that k f (n−k) has a maximum when k = 2 if n > 6. Therefore, the final answer is ⎧ 2 23 ⎪ ⎨ ··· 2 , for odd n > 3, f (n) = 3 23 ⎪ · ⎩ ·· 2 , for even n > 4. Problem 2.43. Consider a set M with m elements and A1 , . . . , An distinct subsets of M such that card(Ai ∩ Aj ) = r ≥ 1 for all 1 ≤ i = j ≤ n. Prove that n ≤ m. Solution 2.43. Observe first that if A is an m×n matrix and m > n, then det A·A = 0, where A denotes the transpose matrix. In fact, let us border the matrix A by a null matrix of size (m − n) × n, in order to get an matrix A . It is obvious that m × m A · A = A · A , and thus det A · A = det A · A = 0. Let D be the n × n matrix whose entries are ' r ≥ 1, if i = j, Dij = dii ≥ r, otherwise. Assume that there exists at most one i such that dii = r. We claim then that det(D) = 0. This follows by subtracting the last line from the first n − 1 and an inductive argument. Consider now the m × n matrix A whose entries are ' 1, if i ∈ Aj , Aij = 0, otherwise. It is now immediate that D = A · A is given by Dij =

n

aik akj = card(Ai ∩ Aj ).

j =1

By hypothesis, Dij = r when i = j , and since the Ai are distinct, there exists at most one i such that Dii = r. The claim from above tells us that det(D) = det A · A = 0, while the first observation implies that m ≤ n. Problem 2.44. Set π(n) for the number of prime numbers less than or equal to n. Prove that there are at most π(n) numbers 1 < a1 < · · · < ak ≤ n with gcd(ai , aj ) = 1.

112

6 Algebra and Combinatorics Solutions

Solution 2.44. If X is a set, we denote by P(X) the set of nonempty subsets of X. Define the map h : {1, . . . , n} → P({p1 , . . . , pπ(n) }) that associates to the integer x the set of prime divisors of x. One then has card(h(ai )) ≥ 1 and further h(ai )∩h(aj ) = ∅, from our assumption. Since the maximal number of disjoint subsets of {p1 , . . . , pπ(n) } is bounded by π(n), the claim follows. Problem 2.45. Prove that for every k, there exists n such that the nth term of the Fibonacci sequence Fn is divisible by k. Recall that Fn is determined by the recurrence Fn+2 = Fn + Fn+1 , for n ≥ 0, where the first terms are F0 = 0, F1 = 1. Solution 2.45. Using the pigeonhole principle, there exists an infinite sequence ni such that Fn1 ≡ Fn2 ≡ · · · ≡ Fnm ≡ · · · (mod k). Moreover, there exists an infinite subsequence nis such that Fni1 +1 ≡ Fni2 +1 ≡ · · · ≡ Fnim +1 ≡ · · · (mod k) by the same argument. Thus one knows that Fnis ≡ Fnit (mod k) and Fnis +1 ≡ Fnit +1 (mod k). The recurrence relation of the Fibonacci numbers shows that Fnis +m ≡ Fnit +m (mod k), for any m ∈ Z. Since F0 = 0, we find that there exists n (actually infinitely many such integers) such that k divides Fn . Problem 2.46. Consider a set of consecutive integers C + 1, C + 2, . . . , C + n, where C > nn−1 . Show that there exist distinct prime numbers p1 , p2 , . . . , pn such that C + j is divisible by pj . Solution 2.46. Let k ≤ h and consider the factorization of C + k into prime factors. If C + k has at least n distinct prime divisors, then we can find one prime divisor that is not yet associated with the other n − 1 numbers C + i. Assume then that C + k has at most j ≤ n − 1 prime factors, which are p1k , . . . , pj k for 1 ≤ j ≤ n − 1. We write down the factorization as a

a1k · · · pj jkk = C + k > C > nn−1 . p1k a

The pigeonhole principle implies that there exists some prime power qk = pj jkk such that qk > n. We then associate the prime number pj k to C + k. Let us show that this procedure yields distinct prime numbers. Suppose that there exist j and k such that the associated numbers are the same. We know that qk divides C + k and qj divides C + j , and they are both prime powers of the same prime. Since qk > n and qj > n, we obtain gcd(qk , qj ) > n. Moreover, gcd(qk , qj ) divides both C + k and C + j and hence their difference |j − k| < n, which is a contradiction. This proves the claim.

6.2 Algebraic Combinatorics

113

Problem 2.47. Let p be a prime number and f (p) the smallest integer for which there exists a partition of the set {2, 3, . . . , p} into f (p) classes such that whenever a1 , . . . , ak belong to the same class of the partition, the equation k

xi ai = p

i=1

does not have solutions in nonnegative integers. Estimate f (p). Solution 2.47. 1. Suppose that a1 , a2 are in the same class and gcd(a1 , a2 ) = 1. One knows then that any natural number greater than or equal to a1 a2 can be written as √ x1 a1 + x2 a2 , where xi ∈ Z+ . Therefore, the prime numbers smaller than p should to different classes of our partition, and hence √ 1 2 p f (p) > π p 2 ∼ , log p where π(n) denotes the number of primes smaller than n. + , let us consider the sets of consecutive elements At = ( 2. If t ∈ Z p ) p . We claim that the equation ki=1 ai xi = p has no integer t+1 + 1, . . . , t solutions if ai ∈ At for 1 ≤ i ≤ k. In fact, if we had a solution xi , then we would have k k k p p xi < ai xi < xi , t +1 t i=1

k

i=1

i=1

and so t < i=1 xi < t + 1, which is false, since xi are integers. Let us consider the classes A1 , A2 , . . . , AL−1 , where L is an integer to be fixed later. Consider now, for every prime q < p/L, the classes Bq = {q, 2q, . . . , αq, . . .}. It is immediate that the associated linear equation has no solutions if the coefficients belong to some Bq , since q does not divide p. Further, A1 , . . . , AL−1 , B2 , B3 , . . . , B p L form a partition of {2, 3, . . . , p}. The total number of classes of this partition is then L − 1 + π(p/L). " " Setting L =

2p log p ,

we obtain f (p)

0, one can use finitely many foldings of the paper along its sides in either 2 equal parts or 3 equal parts to obtain a rectangle whose sides are in ratio r for some r satisfying 1 − ≤ r ≤ 1 + . Solution 3.19. Observe that log3 2 ∈ Q and thus the residues modulo 1 of the elements n log3 2, with n ∈ Z+ , form a dense subset of (0, 1). This implies that the set of numbers of the form 2a 3b , with a, b ∈ Z, are dense in R. Problem 3.20. Show that there exist at most√three points on the unit disk with the distance between any two being greater than 2.

7.2 Combinatorial Geometry

143

Solution 3.20. Consider the points Ai on the unit√disk of radius 1 and center O. Then |OAi | ≤ 1, and if we assume that |Ai Aj | > 2, then 2 < |OAi |2 + |OAj |2 − π 2|OAi | · |OAj | cos A i OAj . Thus, cos A i OAj < 0 and hence A i OAj > 2 , which implies that n ≤ 3. Comments 56 H.S.M. Coxeter proposed this problem in 1933. The same argument the unit disk, then there exists a pair of points shows that if we have k points Ai within such that |Ai Aj | ≤ max 1, 2 sin πk , and this is sharp for 2 ≤ k ≤ 7. The result was generalized to higher dimensions by Davenport, Hajós, and, independently, Rankin, as follows: There are at most n + 1 points√in the unit ball of Rn such that the distance between any two points is greater than 2. Here is a proof sketch. Let Ai denote the m points satisfying the hypothesis, so that OAi they are different from the origin O. Then we have m ≥ n + 2 unit vectors |OA i| n in R with the property that vi , vj < 0. We claim next that there exist pairwise orthogonal subspaces W1 , W2 , . . . , Wm−n ⊂ Rn that span Rn such that each space Wj contains 1 + dim Wj vectors among the unit vectors above, v1 , . . . , vm , that span Wj . Since m − n ≥ 2, there exist two vectors among the vi that are orthogonal to each other, which contradicts our assumptions. The claim follows by induction on the dimension n. If n = 2, it is immediate. Now, if vi = −vn , for some i < n, then all other vj should be orthogonal to vn (and to vi ). Otherwise, we can write for all i < n, vi = ui + xi vn , for some scalars xi < 0 and some nonzero vector ui orthogonal to vn . This implies that ui , uj < 0, when i = j . We can apply therefore the induction hypothesis for the subspace orthogonal to vn and the set of vectors vi , i < n, and obtain the subspaces Wj . We define then Wi = Wi , for i ≥ 2, and let W1 be the span of W1 and vn . This proves the claim. n The claim shows that if we have √ m ≥ n + 2 points on the unit sphere in R whose pairwise distances are√at least 2, then m ≤ 2n, and moreover, there exist two points at distance precisely 2.

• •

H.S.M. Coxeter: Amer. Math. Monthly 40 (1933), 192–193. R.A. Rankin: The closest packing of spherical caps in n dimensions, Proc. Glasgow Math.Assoc. 2 (1955), 139–144.

Problem 3.21. A convex polygon with 2n sides has at least n diagonals not parallel to any of its sides. The equality is attained for the regular polygon. Solution 3.21. Let l be a side of the polygon. Then we have at most n − 2 diagonals parallel to l. In fact, let d1 , d2 , . . . , dp be these diagonals. Then the pair of segments (d1 , l) determines a quadrilateral (actually a trapezoid) having three common sides with the polygon. Moreover, all the pairs (d2 , d1 ), . . . , (dp−1 , dp ) determine quadrilaterals, each one containing two sides of the initial polygon. This implies that p ≤ n − 2. The total number of diagonals that are parallel to at least one side is at most 2n(n − 2). But there are 2n(2n−3) diagonals, and hence n(2n − 3) − 2n(n − 2) = n 2 diagonals with the desired property.

144

7 Geometry Solutions

Problem 3.22. Let d be the sum of the lengths of the diagonals of a convex polygon P1 · · · Pn and p its perimeter. Prove that for n ≥ 4, we have n n + 1 d n−3 · · · > αn−2 > αn−1 = p. Let us show first that p < α2 . Let Ai Ai+2 ∩ Ai+1 Ai+3 = {Oi+1 }. Observe that |Ai+1 Oi+1 | ≤ |Ai+1 Oi+2 |, since Ai+1 Ai+2 Oi+1 ≤ Ai+1 Ai+2 Oi+2 . The triangle inequality in Ai Oi Ai+1 further yields |Ai+1 Oi+1 | + |Oi+1 Ai+2 | > |Ai+1 Ai+2 |.

Summing up these inequalities, and using the remark, it follows that p < α2 , as claimed. Let now k < n2 − 1, and set Ai Ai+k ∩ Ai+1 Ai+k+1 = {O}. The triangle inequality again shows that |Ai Ai+k | + |Ai+1 Ai+k+1 | = |Ai O| + |OAi+k | + |Ai+1 O| + |OAi+k+1 | > |Ai+1 Ai+k | + |Ai Ai+k+1 |. By summing up over i, we obtain 2αk > αk+1 + αk−1 ; thus αk+1 − αk < αk − αk−1 < · · · < α2 − α1 , and hence the claim. Moreover, if n = 2m, then 2αm > αm+1 + αm−1 = 2αm+1 , and so αm > αm−1 . Thus, we have proved that αk+1 > αk for all k. By summing up all these inequalities over k, we find that 2d > (n − 3)p.

7.2 Combinatorial Geometry

145

Further, consider the triangle inequality |Ai Aj | < |Ai Ai+1 | + · · · + |Aj −1 Aj | and sum up over all i and j ; this yields

n n+1 2d < − 2 p. 2 2 Problem 3.23. Find the convex polygons with the property that the function D(p), which is the sum of the distances from an interior point p to the sides of the polygon, does not depend on p. Solution 3.23. Let K be our polygon, p an interior point, a1 , . . . , an the distances from p to the sides of K, and let u1 , . . . , un be unit vectors perpendicular to the sides. Choose another point, say q, in the interior of K. Denote by v the vector pq. The distance from q to the side with label i is therefore di = ai + ui , v, where , is the scalar product. It follows that D(q) = D(p) +

- n

. ui , v .

i=1

Therefore, a necessary and sufficient condition for D(p) to be constant is that n

ui = 0.

i=1

Observe that the solution can be extended to convex polyhedra in Rm . Problem 3.24. Prove that a sphere of diameter 1 cannot be covered by n strips of width li if ni=1 li < 1. Prove that a circle of diameter 1 cannot be covered by n strips of width li if ni=1 li < 1. Solution 3.24. 1. The lateral area of the body obtained by cutting off the sphere with two planes at distance h is 2πRh. Let us assume that we have a covering by means of the strips Bi . Thenthe sum of areas cut open by the strips Bi is at least the area of the sphere, hence π ni=1 li ≥ π , contradicting our assumptions. 2. If we had such a covering, let Bi∗ be the spatial strip that cuts the planar strip Bi , orthogonal to the plane of the circle. Then the union of the strips Bi∗ would cover the cylinder over the disk, in particular the sphere of diameter 1. This would contradict the first part. Problem 3.25. Consider n points lying on the unit sphere. Prove that the sum of the squares of the lengths of all segments determined by the n points is less than n2 . Solution 3.25. Let O be the center of the sphere and let Xi denote the given points. −−→ We denote by vj the vector OXj , and by , the Euclidean inner product. We have then |Xi Xj |2 = vj − vi , vj − vi .

146

7 Geometry Solutions

Set S = 2S =

i,j

n

|Xi Xj |2 . We have then

vi − v1 , vi − v1 +

i=1

n i=1

= 2n(|v1 |2 + · · · + |vn |2 ) − 2

vi − v2 , vi − v2 + · · · +

n

vi − vn , vi − vn

i=1

vj , vi

i,j

= 2n

n

|vi |2 − 2v, v = 2n2 − 2|v|2 ≤ 2n2 ,

i=1

where v =

n

i=1 vi .

−−→ −−→ Problem 3.26. The sum of the vectors OA1 , . . . , OAn is zero, and the sum of their lengths is d. Prove that the perimeter of the polygon A1 . . . An is greater than 4d/n. −−→ −→ Solution 3.26. If A is an arbitrary point in the plane, then nOA = ni=1 (OAi + n −−→ −−→ Ai A) = i=1 Ai A, which implies that n|OA| ≤ |Ai A|. Let us put A = Ai for i = 1, 2, . . . , n and sum up the inequalities so obtained: d = |OA1 | + · · · + |OAn | ≤

2 |Ai Aj |. n i,j

Using the triangle inequality, |Ai Aj | < |Ai Ai+1 | + · · · + |Aj −1 Aj |, we infer that d≤

2 n−1 n2 − 1 p + 2p + · · · + p = p. n 2 4n

Therefore p ≥ 4d n2n−1 >

4d n .

Problem 3.27. Find the largest numbers ak , for 1 ≤ k ≤ 7, with the property that for any point P lying in the unit cube with vertices A1 . . . A8 , at least k among the distances |P Aj | to the vertices are greater or equal than ak . Solution 3.27. At least eight distances are greater than or equal to 0, and this is sharp by taking P = Aj , and hence a8 = 0. By taking P to be the midpoint of an edge, we get a7 ≤ 21 . Assume that a7 < 21 . Then there exists some P for which at least two distances |P Aj | are strictly smaller than 21 , which is obviously false, since vertices are distance 1 apart. Thus a7 = 21 . √

Consider P as the center of some face of the cube. We obtain then a6 ≤ a5 ≤ 22 . If equality does not hold above, then there exists P for which three distances are √ 2 smaller than 2 . However, given three vertices of the cube, there are at least two of √ them at distance greater than or equal to the diagonal of a face, namely 2. Thus our √ assumption fails and so a6 = a5 = 22 .

7.2 Combinatorial Geometry

Now let P be the center of the cube. We derive that a4 ≤ a3 ≤ a2 ≤ a1 ≤ √

Assume that a4 < √ 3 2 .

3 2 .

147 √

3 2 .

Then there exist P and at least five distances strictly smaller

However, for any choice of five vertices of a cube, one finds a pair of than √ opposite vertices that are distance 3 apart. This contradicts our assumptions and √ thus a4 = a3 = a2 = a1 = 23 . Problem 3.28. The line determined by two points is said to be admissible if its slope is equal to 0, 1, −1, or ∞. What is the maximum number of admissible lines determined by n points in the plane? Solution 3.28. We join two points by an edge if the line determined by them is admissible. We obtain a graph with n vertices. By hypothesis, each vertex has degree at most 4. Since the sum of degrees over all vertices is twice the number of edges, we find that there are at most 2n edges, thus at most 2n admissible lines. Moreover, there are only four directions in the plane, and the same computation shows that there are at most 21 n edges associated to every direction. This implies that for odd n, the maximum number of admissible lines is 2n − 2. It is easy to construct examples in which we have precisely 2n admissible lines if n is even and n ≥ 12. For odd n, the associated graph is obtained by adding an isolated vertex to a maximal graph of order n − 1, and this works for any n ≥ 3. Finally, if f (n) denotes the maximum number of admissible lines, then explicit inspection shows that f (2) = 1, f (3) = 3, and f (4) = 6, and so the upper bound 2n is not attained. It can be shown that f (6) = 11 and f (10) = 19. Therefore, for all n but those from {2, 3, 4, 6, 10}, we have f (n) = 4 n2 . Comments 57 More generally, assume that the admissible directions are k in number. The same proof shows that there therefore exist at most kn 2 admissible lines determined by n points. Moreover, the minimum number of directions (not necessarily admissible) defined by n points is at least n − 1. See also the article: •

P.R. Scott: On sets of direction determined by n points, Amer. Math. Monthly 77 (1970), 502–505.

Problem 3.29. If A = {z1 , . . . , zn } ⊂ C, then there exists a subset B ⊂ A such that n ≥ π −1 z |zi |. z∈B

i=1

Solution 3.29. We reorder the numbers z1 , . . . , zn , −(z1 + · · · + zn ) in a sequence w1 , . . . , wn+1 such that the sequence arg w1 , . . . , arg wn+1 is monotonic. In the complex plane, the points w1 , w1 + w2 , . . . , w1 + · · · + wn+1 , are the vertices of a convex polygon P . Set d(P ) for the diameter of P . Since the diameter of a polygon is the distance between two vertices, there exists B ⊂ A such that d(P ) = zi . zi ∈B

148

7 Geometry Solutions

Let p(P ) be the perimeter of P . We have then p(P ) =

|wi | =

n

n |zi | + zi .

i

i

The main isoperimetric inequality states that for any convex planar polygon Q we have the inequality π d(Q) ≥ p(Q). Applying this to the polygon P settles the claim. Comments 58 Using a refined isoperimetric inequality for polygons with a given number of vertices, one finds that there exists B ⊂ {1, . . . , n} such that n sin(kπ/(2k + 1)) ≥ z |zi |. i (2k + 1) sin(π/2k + 1) i∈B

i=1

Problem 3.30. Let A1 , . . . , An be the vertices of a regular n-gon inscribed in a circle of center O. Let B be a point on the arc of circle A1 An and set θ = A n OB. If we set ak = |BAk |, then find the sum n (−1)k ak k=1

in terms of θ . Solution 3.30. We have ak = 2r sin n

(−1)k ak cos

k=1

kπ n

−

θ 2

. Then

θ θ (2k + 1)π π (2k − 1)π = (−1)k r sin − + sin − 2n 2n 2 2n 2 π θ = −1 + (−1)n+1 r sin − , 2n 2

which leads us to n k=1

' (−1) ak = k

π π −1 cos 2n , for even n, 2r sin θ2 − 2n 0, for odd n.

Problem 3.31. Consider n distinct complex numbers zi ∈ C such that min |zi − zj | ≥ max |zi |. i=j

i≤n

What is the greatest possible value of n? Solution 3.31. Let |zn | = maxi≤n |zi |; then the zi belong to the disk of radius R = |zn | centered at the origin. Moreover, the distance between two points of complex coordinates zi and zj is given by |zi − zj |, which by hypothesis is greater than R. In particular, we have n points whose pairwise distances are at least R within the circle of radius R. Let O be the origin and Zi the points. We order the points Zi in

7.2 Combinatorial Geometry

149

increasing order with respect to their arguments. If the origin is among the Zi , then we choose it to be the first point. Further, there remain at least n − 1 points distinct from the origin. One knows that n

Zi OZi+1 = 2π,

i=2 2π OZi+1 ≤ n−1 . In particular, if n ≥ 8, and thus there exists at least one angle Zi " then |Zi Zi+1 | ≤ R 2 − 2 cos Zi OZi+1 < R, contradicting our assumptions. Thus n ≤ 7 and the maximal value n = 7 is reached by the configuration of six vertices of a regular hexagon together with the origin.

Problem 3.32. The interior of a triangle can be tiled by n ≥ 9 pentagonal convex surfaces. What is the minimal value of n such that a triangle can be tiled by n hexagonal strictly convex surfaces? Solution 3.32. There is an obvious recurrence, showing how to go from k to k + 3.

We still have to find the appropriate tiling for n ∈ {9, 10, 11}. If n = 9, then we have the following tiling of the triangle:

A slight deformation of the left corner pentagon shows that a concave quadrilateral can also be tiled with nine pentagons. Moreover, a tiling by n pentagons of a concave quadrilateral can be adjusted as below to a tiling of a triangle with n + 1 pentagons. This method furnishes a tiling with any number n ≥ 9 of pentagons.

150

7 Geometry Solutions

2. We will show that there are no hexagonal tilings of triangles with convex tiles. Assume the contrary. We call a tiling good if no hexagon tile has sides in common with (at least) two sides of the triangle. Given an arbitrary tiling, we can construct a good tiling of a triangle as follows. First, the triangle can be assumed to be a right triangle by means of the following transformation. Consider the cone in R3 over the initial tiled triangle with a cone vertex far away, and cut it by a convenient plane so that one of the angles of the section become right angled. The trace of the coned tiling in the planar section is still a hexagonal tiling with convex polygons. Further, we double the right triangle along one of its sides and obtain a tiling that has no hexagons with common edges with both of the sides opposite to the doubled side. Applying again the same procedure yields a good tiling. The main feature of a good tiling by convex polygons is that any hexagon having a common side with the triangle (called a boundary hexagon) has precisely one side in common, excepting the case in which the hexagon might have a common angle with the triangle (called an exceptional boundary hexagon). In the last case, we have a vertex of the hexagon and its incident sides that are common with the vertex of a triangle and its incident sides. Thus there are λ ∈ {0, 1, 2, 3} exceptional boundary hexagons. Let a∂ be the total number of boundary hexagons and ain the hexagons that are interior, i.e., they have no sides in common with the triangle.

Construct the dual graph associated to the tiling. This is a planar graph whose vertices have degree 6. The edges going outside of the triangle will be called exterior edges of the graph, and the remaining ones interior edges. The number of vertices of this graph is a = ain + a∂ . Counting the sum of degrees of all vertices, we obtain 6a on one side, and 2ein + eo , where ein (respectively eo ) is the number of interior (respectively exterior) edges. Moreover, each exterior edge comes from a unique boundary hexagon, and two different edges come from different hexagons, excepting the exceptional hexagons, which have two outgoing edges. Thus eo = a∂ + λ. This implies that ein = 3a − 21 (a∂ + λ). If we erase the exterior edges of our graph, then we obtain a planar polygon partitioned into polygons (duals of the hexagons). The total number of vertices is a, the number of edges is ein and the number of faces f can be estimated from above as follows. We have ain vertices in the interior of a polygon with a∂ sides. Each polygon has at least three sides (because our initial hexagons were supposed to be

7.2 Combinatorial Geometry

151

strictly convex). Consider the sum of angles of all polygons involved. On the one hand this yields 2πain + π(aδ − 2), and on the other hand this is at least πf . Thus f ≤ 2ain + a∂ − 2. The Euler–Poincaré theorem requires that a∂ 1 λ 1 = a − ein + f ≤ a − 3a − (a∂ + λ) + 2ain + a∂ − 2 = − 2 − . 2 2 2 But λ ≤ 3 and a∂ ≥ 0 lead us to a contradiction. Comments 59 The same proof shows that there does not exist a tiling of the quadrilateral by strictly convex hexagons. Problem 3.33. We say that a transformation of the plane is a congruence if it preserves the length of segments. Two subsets are congruent if there exists a congruence sending one subset onto the other. Show that the unit disk cannot be partitioned into two congruent subsets. Solution 3.33. Let us assume that the disk D can be partitioned as D = A ∪ B, where A and B are congruent. Let p denote the center of D, and assume that p ∈ A. Let F : A → B be a congruence.

Let rs be the diameter that is perpendicular to the line pF (p). For any point x ∈ D, we have |px| ≤ 1. Thus |F (p)F (x)| ≤ 1, for any x ∈ A, and thus |F (p)y| ≤ 1, for any y ∈ B. Now |F (p)r| > 1 and |F (p)s| > 1, and thus r, s ∈ A. Further, |F (r)F (s)| = |rs| = 2 and thus F (r)F (s) is a diameter of the disk. Since |F (r)F (p)| = |F (s)F (p)| = 1, we derive that F (p) is the midpoint of the diameter F (r)F (s), and thus it coincides with the center p. This contradiction proves the claim. Comments 60 Actually, the same argument shows that the disk cannot be partitioned into finitely many congruent sets. The problem is due to B.L. Van der Waerden. More generally, D. Puppe showed that the same holds for any convex compact set in the plane. See also: •

H. Hadwiger and H. Debrunner: Combinatorial Geometry in the Plane, translated by V. Klee. With a new chapter and other additional material supplied by the translator, New York, 1964.

Problem 3.34. Prove that the unit disk cannot be partitioned into two subsets of diameter strictly smaller than 1, where the diameter of a set is the supremum distance between two of its points.

152

7 Geometry Solutions

Solution 3.34. Observe first that the diameter of a planar set A coincides with the diameter of its closure A in R2 . Assume that the disk is partitioned into two sets A and B of smaller diameter. The points of the boundary circle cannot belong to only one of the partition sets, since opposite points are at distance 1. One the other hand, the unit circle cannot be written as the union of two disjoint nontrivial closed subsets (i.e., closed intervals) because it is connected. This implies that there exists some point M of the unit circle that belongs to both A and B. Consider now the opposite point M on the circle. If M ∈ A, then A has diameter 1; otherwise, M ∈ B and so B has diameter 1, which gives us a contradiction. Problem 3.35. A continuous planar curve L has extremities A and B at distance |AB| = 1. Show that, for any natural number n there exists a chord determined by two points C, D ∈ L that is parallel to AB and whose length |CD| equals n1 . Solution 3.35. Assume that L does not admit chords CD parallel to AB either of length a or of length b. We will show that L does not admit chords CD parallel to AB of length a + b. Let Tx be the planar translation along the line AB of the number of x units. We have, by hypothesis, Ta L ∩ L = ∅ and Tb L ∩ L = ∅. This implies that Ta+b L∩Ta L = ∅. On the other hand, let d(U, V ) denote the distance between the sets U and V in the plane, given by inf P ∈U,Q∈V |P Q|. Then obviously d(Ta+b L, L) ≤ d(Ta L, L) + d(Tb+a L, Ta L). This implies that d(Ta+b L, L) = 0 if d(Ta L, L) = d(Tb+a L, Ta L) = 0. Moreover, L and Ta+b (L) are closed subsets, and thus the distance is attained, i.e., there exist P ∈ L and Q ∈ Ta+b (L) such that |P Q| = d(L, Ta+b (L)). In particular, P = Q and thus Ta+b L ∩ L = ∅, proving our claim. Suppose now that there is no parallel chord of length a = b = n1 . Then we derive that there is no such chord of length n2 . Take a = n1 and b = n2 . The previous claim shows that there is no chord of length n3 . An easy induction shows that there is no chord of length nk for k ≤ n. This contradicts the fact that AB has length 1. Comments 61 The result is due to H. Hopf, who characterized all the possible lengths of chords arising as above. •

H. Hopf: Über die Sehnen ebener Kontinuen und die Schleifen geschlossener Wege, Comment. Math. Helv. 9 (1937), 303–319.

Problem 3.36. The diameter of a set is the supremum of the distance between two of its points. Prove that any planar set of unit diameter can be partitioned into three √ 3 parts of diameter no more than 2 . Solution 3.36. The main ingredient is to show that a planar set F of unit diameter is contained within a regular hexagon whose opposite sides are at distance 1. Given a vector v = v1 , let us consider the thinest strip Sv parallel to v that contains F . Then the strip Sv is bounded by two parallel lines, each line containing

7.2 Combinatorial Geometry

153

at least one point of F , by the minimality of the strip. In particular, the width of the strip is bounded by the distance between two points of F , and so by the diameter of F . Now let v2 be the image of v1 by a rotation of angle π3 , and let v3 be the image of the rotation of angle 2π 3 . The intersection of the corresponding strips Hv = Sv1 ∩Sv2 ∩Sv3 is a semiregular (i.e., all opposite edges are parallel) hexagon having all angles 2π 3 . However, in general, this hexagon is not regular.

We choose two opposite vertices, say A and A , of the parallelogram Sv1 ∩ Sv2 . For instance, we choose A such that the vectors spanning the parallelogram issued from it are positive multiples of both v1 and v2 . We denote by x the distance from A to the strip Sv3 and by x the distance from A to the strip Sv3 . We observe that the hexagon Hv is regular if x = x . Moreover, let us set f (v) = x − x . Obviously, the function that associates to the vector v the quantity x − x is continuous. Let us compute f (−v), by rotating v through π. This amounts to interchanging A with A , and thus f (−v) = −f (v). A continuous function taking both positive and negative values must have a zero. Thus there exists some position of v for which f (v) = 0 and so Hv is regular. Further, the regular hexagon of width 1 can be partitioned into three pieces having √ diameter 23 , as in the figure below:

This proves the claim. Comments 62 K. Borsuk proved in 1933 that a planar set of diameter 1 can be partitioned into subsets of diameter d < 1. The main ingredient above is a lemma due to J. Pál, from 1920. B. Grünbaum proved (improving on ideas of D. Gale) that a set of unit diameter in R3 is contained in a regular octahedron of width 1, with three of its vertices (one vertex on each diagonal) cut off by planes orthogonal to the diagonals,

154

7 Geometry Solutions

at distance 21 from the center. Using a suitable dissection of this polyhedron, one finds that the set F can be partitioned into four pieces having diameter not bigger than / √ 6129030 − 937419 3 ≈ 0.9887. √ 1518 2 Gale’s conjecture from 1953, " that one can always get a partition into four pieces of diameter no bigger than • •

√ 3+ 3 6 ,

is still open.

D. Gale: On inscribing n-dimensional sets in a regular n-simplex, Proc. Amer. Math. Soc. 4 (1953), 222–225. B. Grünbaum: A simple proof of Borsuk’s conjecture in three dimensions, Math. Proc. Cambridge Philos. Soc. 53 (1957), 776–778.

Problem 3.37. 1. Prove that a finite set of n points in R3 of unit diameter, can be 2 covered by a cube of side length 1 − 3n(n−1) . 2. Prove that any planar set of n points having unit diameter can be partitioned into √ 3 2π three parts of diameter less than 2 cos 3n(n−1) . Solution 3.37. 1. A direction will mean below a line without specified orientation. The angle between two directions is the among the four angles they define, and

smallest thus it is always within the range 0, π2 . Consider the set of n points {x1 , x2 , . . . , xn }, which determine at most N = n(n−1) distinct directions xi xj , which we denote 2 by l1 , l2 , . . . , lN . We want to prove first that there exist three mutually orthogonal directions y1 , y2 , y3 , so that 1 ∠(yi , ls ) ≥ arccos 1 − . min 1≤i≤3,1≤s≤N 3N Consider an orthogonal frame in R3 given by three mutually orthogonal vectors e1 , e2 , e3 , and choose the rotation R1 and R2 of the space with the property that R1 e1 = e2 , R1 e3 = e3 , R2 e1 = e3 , R2 e2 = e2 . Using these rotations, we have min

1≤i≤3,1≤s≤N

∠(yi , ls ) = min min(∠(y1 , ls ), ∠(y1 , R1 ls ), ∠(y1 , R2 ls )). 1≤s≤N

Consider the set of 3N directions D = {lj , R1 lj , R2 lj , 1 ≤ j ≤ N }. We will prove the following estimate from below: 1 = θ, max min ∠(y, l) ≥ arccos 1 − y l∈D 3N for an arbitrary set D of 3N directions. Assume the contrary. The set of directions y making an angle smaller than θ with a given direction l form a symmetric cone Cl,θ of angle 2θ with axis l. If there is no direction y satisfying the inequality above, then the union of all cones Cl,θ , over l ∈ D will cover the set of all directions in space. Each symmetric cone Cl,θ intersects the unit sphere along a symmetric pair of

7.2 Combinatorial Geometry

155

spherical caps of radius θ. In particular, these spherical caps cover the surface of the whole sphere, since a point not covered yields a direction not belonging to the union. Furthermore, this implies that the total area of the spherical caps is (strictly) greater than the area of the sphere. Now, the area of each spherical cap is 2π(1 − cos θ), and thus the total area covered by the cones is 12πN (1 − cos θ) = 4π . But 4π is the total area of the sphere, which contradicts the claim. Observe that the inequality must be strict, since N ≥ 1 and any covering of the sphere by nontrivial spherical caps will have some overlaps. Consider now three mutually orthogonal directions yj satisfying the claim above. Let further ξj ⊂ yj be the image of the given set of points under orthogonal projection onto yj . The distance between the images of xi and xk on yj is |xi xk | cos ∠(yj , xi xk ) ≤ sin θ . Thus each subset ξj is contained in some interval of length less than cos θ , and thus the set of points is contained in the strip of width sin θ, orthogonal to yj . The intersection of the three orthogonal strips associated to the direction yj is a cube of side length cos θ , and the claim follows. 2. The second part follows using the same idea. This time, we consider three directions y1 , y2 , y3 in the plane such that y2 is obtained by a counterclockwise rotation R1 of angle π3 from y1 , and y3 by a counterclockwise rotation R2 of angle π3 from y2 . We claim that for any N directions in the plane, we have min

1≤i≤3,1≤s≤N

∠(yi , ls ) ≥

π = α. 6N

The same trick as above shows that the 3N symmetric angle sectors of angle 2α cover the set of all directions and thus the unit circle. Now, the total length of arcs covered by these symmetric cones is 12N α = 2π , and the claim follows. Therefore, the is contained within the intersection of three strips of set of points 2π width w = cos 3n(n−1) , which are orthogonal, respectively, to the directions yj . The intersection is a semiregular hexagon H . Since all three strips have the same width, we find that there are only two different side lengths, as in the figure below:

This means that H is the intersection of two equilateral triangles of the same incenter O. Let A1 , . . . , A6 be the vertices of H . We have then |A1 A2 | = |A3 A4 | = |A5 A6 | = a and |A2 A3 | = |A4 A5 | = |A6 A1 | = b. Moreover, by computing the

156

7 Geometry Solutions

width of H , we find that the lengths a, b are constrained to satisfy a + b = √2 w. 3 Assume from now on that a ≤ b. We divide the hexagon H by means of three segments OP , OQ, OR orthogonal to the three longest sides of the hexagon. The diameter of the pentagon √ OP A1 A2 Q is one of its diagonals, and one computes easily |P Q| = a+ b 2,

|OA1 | =

3b2 +(2a+b)2 √ , 2 3

and |P A2 | ≤ |P Q| since ∠(P A2 Q) ≥ ∠(P QA2 ) = √ 3 2 w,

π 3.

Using the relation between a and b, we obtain that |P Q| ≤ 43 (a + b) ≤ with equality only when a = b and the hexagon is regular. On the other hand, |OA1 | ≤ 2w 3 , with equality when a = 0. This implies that the diameter of each piece is at most √ 3 2 w, where w has the value found above. Comments 63 This version of Borsuk’s problem for finite sets and its higherdimensional generalization for covering by cubes was given in: •

L. Funar: A minimax problem in geometry (in Romanian), Proc. 1986 Nat. Conf. Geometry and Topology - Tîrgoviste, Univ. Bucharest, 91–96, 1988.

Problem 3.38. Prove that any convex body in Rn of unit diameter having a smooth boundary can be partitioned into n + 1 parts of diameter d < 1. Solution 3.38. The first step is to observe that the claim holds for the unit ball B n . We consider the regular spherical simplex obtained by projecting the regular simplex onto the sphere from its incenter. The n + 1 faces of the spherical simplex form a dissection into pieces of diameter smaller than 1, since there are no opposite points lying in the same face. Otherwise, one can compute explicitly the diameter of each piece. Consider the convex body F with smooth boundary W ⊂ Rn . The only point where the smoothness is used is when we assume that there exists a unique tangent hyperplane Tp W at each point p of W . Moreover, there exists a unique unit vector vp in Rn , based at p, that is orthogonal to the tangent hyperplane Tp W and points toward the half-space that does not contain F . This permits us to define a smooth map G : W → S n−1 to the unit sphere centered at the origin O, usually called the Gauss map in differential geometry. We associate to the point p ∈ W the unique point G(p) ∈ S n−1 for which the vectors OG(p) and vp are parallel, and so the tangent space Tp W is parallel to the tangent hyperplane TG(p) S n−1 at the point G(p) on the sphere. Since F is convex and W is smooth, the map G is a smooth diffeomorphism. Consider now the preimage by G of a partition of the unit sphere into pieces of diameter smaller than 1. This yields a partition of W into n + 1 pieces, which we denote by A1 , A2 , . . . , An+1 . We claim that each piece Aj has diameter smaller than 1. Assume the contrary. We can replace the Aj by their closures, without modifying their diameters. Now there should exist two points x, y ∈ Aj at distance 1, which is the diameter of F . Let Tx W and Ty W be the tangent planes at W at x, y respectively. Then Tx W and Ty W are orthogonal to the segment xy. Otherwise, a small perturbation x ∈ W of x in some tangent direction from Tx W , making an obtuse angle with xy, will increase the length of |x y|, but the diameter of W is no larger than 1.

7.2 Combinatorial Geometry

157

Therefore G(x) and G(y) are opposite points on the unit sphere, and moreover, they belong to the same set of the partition, but this is impossible, since the pieces of the partition are of diameter strictly smaller than 1. Comments 64 The question whether any set F in Rn can be partitioned into n + 1 pieces of strictly smaller diameter is known as Borsuk’s problem. Since the diameter of a set equals the diameter of its convex hull, it suffices to consider the problem the convex sets. The result in the problem above was obtained by H. Hadwiger in 1945. The case left open is that in which the convex body has corners. There was no significant progress in the subject until 1993, when J. Kahn and G. Kallai gave an unexpected counterexample by showing that the minimal number f (n) of pieces√needed for a dissection into parts of strictly smaller diameter satisfies f (n) ≥ (1.1) n for large n and that f (n) > n + 1 for n ≥ 1825. The subsets that they considered are finite sets of points. This bound was subsequently lowered by N. Alon and then, by A. Hinrichs, and C. Richter to n = 298. A thorough treatment of the subject is in the recent book of Boltyanski, Martini, and Soltan. However, the 4-dimensional case is still open. • • • •

V. Boltyanski, H. Martini, and P.S. Soltan: Excursions into Combinatorial Geometry, Springer, Universitext, 1997. H. Hadwiger: Überdeckung einer Menge durch Mengen kleineren Durchmesser, Commentarii Math. Helvet. 18 (1945/1946), 73–75, 19 (1946/1947), 71–73. A. Nilli: On Borsuk’s problem, Jerusalem combinatorics ’93, 209–210, Contemp. Math., 178, Amer. Math. Soc., Providence, RI, 1994. J. Kahn and G. Kalai: A counterexample to Borsuk’s conjecture, Bull. Amer. Math. Soc. 29 (1993), 1, 60–62.

Problem 3.39. Let D be a convex body in R3 and let σ (D) = supπ area(π ∩ D), where the supremum is taken over all positions of the variable plane π . Prove that D can be divided into two parts D1 and D2 such that σ (Di ) < σ (D). Solution 3.39. Choose a point in the interior of D and a sequence of chords l1 , l2 , . . . , ln , . . . of D passing through this point, the chords being dense in D. Choose also a sequence ε of positive numbers decreasing rapidly to zero. Let Cn be the set of points at distance less than εn from ln . By choosing ε small enough, we can ensure ∞ −n that ∞σ (Cn ) < 2 σ (D). Set D1 = ∪n=1 Cn and D2 = D − D1 . Then σ (D1 ) < n=1 σ (Cn ) < σ (D). It is also clear that for any plane π , area(π ∩ D2 ) < σ (D). But D2 is a compact and area(π ∩ D2 ) is a continuous function on the plane π . The set of planes that intersect D can be parameterized by a point in D (which will be considered the origin) times the projective space RP 2 , and thus by a compact set. Thus area(π ∩ D2 ) achieves its minimum, which will therefore be strictly smaller than σ (D). Comments 65 The problem was proposed by L. Funar and the solution above was given by C.A. Rogers. A more interesting question is to consider dissections into convex subsets. The additional convexity assumption implies that the boundary structure of

158

7 Geometry Solutions

each piece (at the interior of the body) should be polyhedral. In particular, the minimal number of convex pieces with smaller σ , which the unit ball requires, is 3. If the body D is strictly convex (meaning that the open segment determined by two of its boundary points is contained in the interior of D) or D is a polyhedron without parallel edges, then we can show that four pieces suffice. By a recent result of M. Meyer, two plane sections of maximal area cannot be disjoint. Fix a maximal-area cross-section F and let F (ε) be the set of points of D, at distance less than ε from F . The complement of F (ε) has two components, A+ and A− , which are of strictly smaller σ , because there is no maximal-area plane section contained completely in D − F (ε). Further, F (ε) can be subdivided, by a hyperplane orthogonal to F , into two pieces of smaller σ if ε is small enough. However, we don’t know whether three pieces suffice for an arbitrary convex body. A related problem is whether we have σ (D) ≤ 41 area(∂D), where ∂D denotes the boundary of D. This would imply that the number of maximal area planar sections with disjoint interiors is at most 4, with equality for the regular simplex. • •

L. Funar: Problem E 3094, Amer. Math. Monthly 92 (1985), 427. M. Meyer: Two maximal volume hyperplane sections of a convex body generally intersect, Period. Math. Hungar. 36 (1998), 191–197.

Problem 3.40. If we have k vectors v1 , v2 , . . . , vk in Rn and k ≤ n + 1, then there 1 exist two vectors making an angle θ with cos θ ≥ − k−1 . Equality holds only when the endpoints of the vectors form a regular (k − 1)-simplex. Solution 3.40. Assume that the vectors vj are unit vectors. Then ⎞ ⎛ 2 k k 2 2 |vi − vj | ≤ ⎝ |vi − vj | ⎠ + vi = k |vi |2 = k 2 . 1≤i<j ≤k

1≤i<j ≤k

i=1

i=1

2k Thus there exists at least one pair i, j for which |vi − vj |≤ k−1 . This is equivalent to the fact that the angle θ between vi and vj satisfies the claimed inequality.

Problem 3.41. 1. Consider a finite family of bounded closed convex sets in the plane such that any three members of the family have nonempty intersection. Prove that the intersection of all members of the family is nonempty. " 2. A set of unit diameter in R2 can be covered by a ball of radius

1 3.

Solution 3.41. 1. Assume that the convex sets are A1 , A2 , . . . , Ak and k ≥ 4. We claim that if each k − 1 sets among them have a common point, then all of them should have a common point. The result of the problem follows by using recurrently the previous claim. Before proceeding, we observe that given a set of k ≥ 4 points in the plane, we can find two subsets whose convex hulls intersect nontrivially. Here the convex hull of a set of points is the smallest convex polygon containing these points (and thus all

7.2 Combinatorial Geometry

159

segments pairwise connecting them). In fact, if we choose four points among them, then their convex hull is either a triangle containing the fourth point in its interior, or else a quadrilateral having two diagonals crossing each other. Let pi be a point belonging to ∩j =i,j ≤k Ai and possibly not to Ai . There exists therefore a partition of the set {p1 , p2 , . . . , pk } = P ∪ Q, where P = {pi1 , pi2 , . . . , pim } and Q = {pj1 , pj2 , . . . , pjk−m }, such that the convex hull of the subset P intersects in at least one point the convex hull of the subset Q. Recall that each point of P belongs to ∩i∈{i1 ,i2 ,...,im } Ai , which is convex as the intersection of convex sets. Thus the convex hull of P is equally contained within ∩i∈{i1 ,i2 ,...,im } Ai . The same argument shows that the convex hull of Q is contained in ∩j ∈{j1 ,j2 ,...,jk−m } Aj . Since the two convex hulls have at least one common point, this therefore belongs to ∩ki=1 Ai , as claimed. " 2. It suffices to see that any finite subset is contained in a disk of radius 31 , which " means that the intersection of balls centered at that points of radius 13 is nonempty. By the previous claim, it suffices to consider the case of three points. If the triangle made by these points is obtuse, then we can take the disk whose diameter is the largest side of the triangle. If not, then all angles of the triangle are acute and there exists " at least one angle, say A ≥

π 3.

The circumradius of the triangle is R =

a 2 sin A

≤

1 3.

Comments 66 The first part of the problem is the particular case of the celebrated Helly’s theorem, proved in 1923: if a family of closed convex sets in Rn has the property that any n + 1 among them have a common point, then the intersection of all sets is nonempty. The second claim was generalized by Jung in 1901, who proved that any set of diameter 1 in Rn admits " an unique n-ball of minimal radius covering it, and the minimal radius n is bounded by 2(n+1) . The original proof of Jung’s theorem was quite complicated, but there exists an elementary proof given later by Blumenthal and Wahlin. It is worthy of notice√that any subset of Rn of diameter 1 can be covered also by a , according to Gale. simplex of diameter n(n+1) 2 •

• • •

L.M. Blumenthal, G.E. Wahlin: On the spherical surface of smallest radius enclosing a bounded subset of n-dimensional Euclidean space, Bull. Amer. Math. Soc. 47 (1941), 771–777. L. Danzer, B. Grünbaum, and V. Klee: Helly’s theorem and its relatives, Convexity, Proc. Symposia Pure Math. vol. 7, AMS, 1963, 101–180. D. Gale: On inscribing n-dimensional sets in a regular n-simplex, Proc. Amer. Math. Soc. 4 (1953), 222–225. H.W.E. Jung: Über die kleinste Kugel, die eine räumliche Figur einschliesst, J. Reine Angew. Math. 123 (1901), 241–257.

Problem 3.42. Let T be a right isosceles triangle. Find the disk D such that the difference between the areas of T ∪ D and T ∩ D is minimal.

160

7 Geometry Solutions

Solution 3.42. Let us show that the circle C bounding D has the following two properties, valid for an arbitrary triangle T : 1. The circle C is split into three arcs outside T and three arcs inside T . Then the total length of the outside arcs equals the total length of the inside arcs. 2. The three chords determined by the edges of T have their lengths proportional to the respective side lengths. 1.Assume that C has fixed center. When the radius of C increases by ε, area(T −D) decreases as lin ε 2 , where lin is the length of the inside arcs. On the other hand, area(D − T ) decreases as lou ε 2 , where lou is the length of the outside arcs. Thus we have a local minimum only if (lin − lou) ε 2 does not increase and thus lin = lou . 2. Assume now that C has fixed radius. Since area(T − D) − area(D − T ) = area(T ) − area(D), it is sufficient to minimize area(D − T ). Consider that the center of D is slightly translated by εu, where u is a unit vector. Then we can compute the change of area(S − T ) by summing the contribution of each chord sector. The contribution of A1 A2 is εu × A1 A2 and thus the total variation is εu × A1 A2 + B1 B2 + C1 C2 . The variation must be zero in all directions u and so we obtain the condition A1 A2 + B1 B2 + C1 C2 = 0. This implies that the chords are proportional to the side lengths.

If T is right isosceles triangle of unit side length, then the center O of the circle C belongs to the altitude, by symmetry considerations. Denote by x the distance from the center O to the edges CB and CA, by y the distance from O to the edge AB. The angles determined by the outside √ chords are 2α, 2β, 2β respectively. Since O belongs to the altitude, we have y + 2x = 21 . Moreover, the first claim above implies that √ r 2 − x 2 and α + 2β = π2 and thus cos α = sin 2β. This translates into yr = 2x r2 √ 2 2 2 2 4 2 2 thus √ y r = 4x r − 4x . The second statement above 2implies that r − y / 2 = r 2 − x 2 , and thus introducing above, we obtain 2r = 4x − 1. In particular, x satisfies the equation 16x 4 − 16x 3 − 12x 2 + 8x − 1 = 4x 2 + 2x − 1 4x 2 − 6x + 1 = 0.

7.2 Combinatorial Geometry

161

√

The only convenient root is x = 5−1 4 , since among the remaining ones, one is negative, another one is larger than 1, and the last one would lead to a negative value for r. Thus 0 √ √ 1 √ 1+ 2− 5 r= 5−2 , y = , √ 2 2 2 and the circle is determined by these parameters. Problem 3.43. Let r be the radius of the incircle of an arbitrary triangle lying in the √ 5−1 closed unit square. Prove that r ≤ 4 . Solution 3.43. It is clear that the inradius of a triangle contained in a larger triangle is not bigger than the inradius of the bigger triangle. Thus it suffices to consider only triangles whose vertices lie on the boundary of the square. Further, at least one vertex of the triangle can be assumed to be in a corner of the triangle. If two vertices of the triangle belong to the same edge, then we can assume that these vertices are the endpoints of that edge, as above. Thus it suffices to consider the triangle OAB whose vertices are O = (0, 0), A = (a, 1), and B = (1, b) with 0 ≤ a, b ≤ 1. The inradius of a triangle of area S and semiperimeter p is given by r = pS . Therefore, r=√

1 + a2

+

√

1 − ab . 1 + b2 + (1 − a)2 + (1 − b)2

The inequality to be proved is then equivalent to √ F (a, b) = 1 + a 2 + 1 + b2 + (1 − a)2 + (1 − b)2 − 5 + 1 (1 − ab) ≥ 0 ∂2F > 0 by a simple computation. ∂2a √ 6 √3 2 . Let ε = 1 − (1+ 5)2 1 1. If 1 ≤ b ≤ 4 , set f (x) = F (x, b). Then we observed above that f (x) is √ 1 > 0, then f (x) > 0 for any increasing, and since f (0) = 1 + 5 y − y 2 −2y+2

for 0 ≤ a, b ≤ 1. We have F (a, 0) ≥ 0. Further,

x, and thus f (x) is increasing. In particular, F (a, b) ≥ F (0, b) ≥ 0, provided that 1 4 ≤ b ≤ 1. 2. If 0 ≤ a ≤ 41 and 0 ≤ y ≤ ε, we have

√ x 1−x − f (x) = 1 + 5 b + √ 1 + x2 (1 − x)2 + (1 − b)2 √ x 1−x . ≤ 1+ 5 ε+ √ −√ 2 2 1+x x − 2x + 2

Thus f ( 41 ) < 0. As f is increasing, we have f (x) < 0, for x ∈ 0, 41 , and so f

is decreasing on 0, 41 . But f 41 = F 41 , b = F b, 41 ≥ 0, by the previous case. Thus F (a, b) ≥ 0, for 0 ≤ a ≤ 41 and 0 ≤ y ≤ ε. 3. Since F is symmetric, it suffices to check now the case ε ≤ a ≤ 41 and ε ≤ b ≤ 41 . Actually, we will use only the fact that ab ≥ ε2 . This amounts to saying

162

7 Geometry Solutions

that the area S of the triangle OAB is S = p√2 , 3 3

1−ab 2

≤

1−ε2 2 .

Any triangle satisfies the

obtained from the Heron formula for the area by the means inequality S ≤ inequality. In our case, we have r=

√ S S ≤ √ ≤ 5 + 1. p 3 3S

This settles our claim. Comments 67 The solution presented here is due to Abi-Khuzam and Barbara: • •

L. Funar: Problem 6477, Amer. Math.Monthly 81 (1984), 588. F. Abi-Khuzam, R. Barbara: A sharp inequality and the inradius conjecture, Math. Inequalities and Appl. 4 (2001), 323–326.

Problem 3.44. Let P be a point in the interior of the tetrahedron ABCD, with the property that |P A| + |P B| + |P C| + |P D| is minimal. Prove that AP B = CP D and that these angles have a common bisector. Solution 3.44. Let fA (x) = |AX|. Then fA : E 3 → E 3 is a smooth function on E 3 − {A}. Moreover, the gradient function (∇fA )(X) is the unit vector of direction AX AX, namely |AX| . Consider now the function g = fA + fB + fC + fD for P ∈ int(ABCD). Then g having a minimum at P implies that (∇g)(P ) = (∇fA )(P ) + (∇fB )(P ) + (∇fC )(P ) + (∇fD )(P ) = 0. If a, b, c, d are the unit vectors of directions AP , BP , CP , DP , then the previous conditions reads a + b + c + d = 0, or alternatively, −(a + b) = c + d. Observe now that (a+b) is the direction of the bisector of the angle AP B, while c+d is the direction of the bisector of CP D. Thus these two angles have a common bisector. On the other hand, we have a + b, a + b = c + d, c + d, and because |a| = |b| = |c| = |d|, by simplifying the terms, we obtain that a, b = c, d. This is equivalent to saying that AP B = CP D. Problem 3.45. Let OA1 , . . . , OAn be n linearly independent vectors of lengths a1 , . . . , an . We construct the parallelepiped H having these vectors as sides. Then consider the n altitudes in H as a new set of vectors and further, construct the parallelepiped E associated with the altitudes. If h is the volume of H and e the volume of E, then prove that he = (a1 . . . an )2 . Solution 3.45. Let vi = OAi and wi be the altitude vectors. We have the relations wj , vi = δij vi 2 , where δij is Kronecker’s symbol and , is the scalar product. The linear functionals fi (vj ) = δij vi 2 are linearly independent, and thus the dual vectors wj are linearly independent. Thus E is well defined.

7.2 Combinatorial Geometry

163

Moreover, let A be the matrix consisting of the components of the vi ’s and let B be the matrix consisting of the components of the wj ’s. It follows that AB = diag v1 2 , . . . , vn 2 , where diag(c1 , . . . , cn ) denotes the diagonal matrix with given entries. Therefore vol(E)vol(H ) = | det A| · | det B | = (a1 . . . an )2 . Problem 3.46. Let F be a symmetric convex body in R3 and let AF,λ denote the family of all sets homothetic to F in the ratio λ that have only boundary points in common with F . Set hF (λ) for the greatest integer k such that AF,λ contains k sets with pairwise disjoint interiors. Prove that hF (λ) ≤

(1 + 2λ)3 − 1 . λ3

Solution 3.46. Set Bλ = ∪H ∈AF,λ H . We prove first that Bλ ⊂ (1 + 2λ)F, where αF denotes the homothety of F in ratio α. If v is a point of the boundary ∂F , let v be given by |ov | = (1 + λ)|ov|, where o is the symmetry center of F . Set Fv for the set in AF,λ having center at v . Let x be a point on the boundary of Fv , ox ∩ ∂F = {a}, ov ∩ ∂Fv = {q}, and let vx be parallel to qx, with x ∈ ∂F . Then x = v, xo + v, ox ≥ v, ox, , qv

so that

x. , = qv v, ox ≤ vox

Since F is convex, the intersection of the two segments |oa| ∩ |vx | is nonempty and contains at least one point, say b. Then |vx | ⊂ F , b ∈ F , b ∈ |ox|. Since oa ob ov 1 ≥ = = , ox ox oq (1 + 2λ) the point x belongs to (1 + 2λ)F . Now assume that we have k subsets Fi ∈ AF,λ that have pairwise disjoint interiors. Then their union is contained in Bλ , and all of them are disjoint from F . This means that their union is contained in (1 + 2λ)F − F . Their total volume cannot therefore exceed vol((1 + 2λ)F − vol(F ), and thus their number is bounded by (1 + 2λ)3 − 1 1 vol(1 + 2λ)F − vol(F ) = . vol(F ) λ3 Comments 68 Hadwiger considered the problem of finding estimates for hF (1) (which is called the Hadwiger number of F ) and proved in 1957 that

164

7 Geometry Solutions

n2 + n ≤ hF (1) ≤ 3n − 1 for a convex body F in Rn . Grünbaum conjectured in 1961 (and proved for n = 2) that for any r satisfying these inequalities there exists a convex body F such that hF (1) = r. More generally, the proof used above shows that hF (λ) ≤

(1 + 2λ)n − 1 . λn

We have equality above only if λ1 ∈ Z+ and F is a parallelohedral body. The result is due to V. Boju and L. Funar. There are no good estimates from below except in dimension 2. In fact, if n = 2, the function λhF (λ) approaches an interesting invariant of the oval F , called the intrinsic perimeter. Define the norm ∗ F by means of xyF =

xy , oz

where z is a point of the boundary ∂F such that xy and oz are parallel. This is called the Minkowski norm (or metric) defined by the oval F . The intrinsic perimeter p(F ) of ∂F is the limit of the perimeters of polygons inscribed in F that approach ∂F . For instance, the intrinsic perimeter of a circle is 2π and that of a square is 8. Golab in 1932 and Reshetnyak in 1953 showed that the intrinisc perimeter satisfies the inequalities 6 ≤ p(F ) ≤ 8 with equality in the left side only when F is a hexagon, and on the right side only when F is a rectangle. Then it was proved by V. Boju and L. Funar that the intrinisc perimeter is related to Hadwiger numbers by means of the formula p(F ) = 2 lim λhF (λ), λ→0

and moreover, we have 3+

3 4 ≤ hF (λ) ≤ 4 + , λ λ

with equality on the left side only if λ1 ∈ Z+ and F is an affine regular hexagon. More about packing and covering invariants of convex bodies can be found in the recent book of Böröckzy, and a complete survey on the Minkowski geometry in the book of Thompson. • • •

V. Boju and L. Funar: Generalized Hadwiger numbers for symmetric ovals, Proc. Amer. Math. Soc. 119 (1993), 931–934. K. Böröczky Jr.: Finite packing and covering, Cambridge Tracts in Mathematics, 154, Cambridge University Press, Cambridge, 2004. H. Hadwiger: Über Treffanzahlen be. translationsgleichen Eikörpern, Archiv Math. 8 (1957), 212–213.

7.2 Combinatorial Geometry

•

165

A.C. Thompson: Minkowski geometry, Encyclopedia of Mathematics and Its Applications, 63. Cambridge University Press, Cambridge, 1996.

Problem 3.47. Let denote the square of equations |xi | ≤ 1, i = 1, 2, in the plane, and let A = (a1 |a2 ) be an arbitrary nonsingular 2 × 2 matrix partitioned into two columns. We identify each column with a vector in R2 . Prove that the following inequality holds: a1 , xa2 , x 1 = , minmax 2 A x∈ det A where x, y = x1 x2 + y1 y2 is the usual scalar product. Solution 3.47. Let f (A, x) be the function under minimax. Taking A0 = −11 11 , we obtain 1 max |f (A0 , x)| = . x∈ 2 Thus it suffices to show that for general A, we have max |f (A, x)| ≥ x∈

1 . 2

This is equivalent to showing that | det A| ≤ 1, provided that max |a1 , xa2 , x| = x∈

1 . 2

Assume that the contrary holds, and so | det A| > 1. The set of points in the plane with coordinates (a1 , x, a2 , x) ∈ R2 , for x ranging in the square , is a parallelogram π centered at the origin O and having area 4| det A| > 4. According to our claim,!the parallelogram π is contained in the planar region Z = (u, v) ∈ R2 ; |uv| ≤ 21 . Let us consider the point P = (0, s), where s > 0, is a boundary point of π, and let L ⊂ R2 be the lattice generated by the vectors (1/s, s/2) and (−1/s, s/2). The lattice has area 21 . Then, according to Minkowski’s theorem, the interior of π contains at least one point of the lattice L, different from the origin. Let this point be given by m(1/s, s/2) + n(−1/s, s/2) = ((m − n)/s, (m + n)s/2), where m, n ∈ Z. We can also suppose gcd(n, m) = 1. Since the interior of π is contained in Z, it follows that |m2 − n2 | < 1 and thus m2 = n2 . The only possibilities are (0, ±s) and (±2/s, 0). The points (0, ±s) are on the boundary of π and they are not convenient. If Q = (2/s, 0) ∈ int(π ), then the midpoint S of the segment |P Q| should also belong to the interior of π . However, S = (1/s, s/2) belongs to the boundary curve of equation |uv| = 1/2 of the region Z containing π , so it cannot lie within the interior of π. This contradiction proves our claim. Problem 3.48. We denote by δ(r) the minimal distance between a lattice point and the circle C(O, r) of radius r centered at the origin O of the coordinate system in the plane. Prove that lim δ(r) = 0. r→∞

166

7 Geometry Solutions

Solution 3.48. Let G be a lattice point on the axis Ox such that |OG| = [r]. Draw a perpendicular line at G to Ox that intersects the circle C(O, r) at the point C situated between the lattice points A and B, in the upper half-plane. We choose B to be the point that lies in the interior of the disk D(O, r) of radius r. Notice that |AB| = 1, because these are neighboring lattice points on the same axis line. Let us now draw BD⊥AB, where D is again a lattice point, |BD| = 1, and D is situated outside the disk D(O, r). Since the problem asks for the value at the limit of δ(r) when r → ∞, we can assume that r > 4. Let now consider the line CF that is tangent to the circle C(O, r) and such that F ∈ BD. We show first that |BF | ≤ |BD|. < CDB ≤ ADB = Assume the contrary, namely that |BF | > |BD|. Then CFB π . Furthermore, 4 √ = sin arcsin [r] > r − 1 > 2 = sin π , if r > 4. sin CF B = sin OCG r r 2 4 Thus CF B > π4 , which is false. Consequently, |BF | ≤ |BD| = 1. The triangles CBF and OCG are similar, and therefore √ √ |BC| r 2 − [r]2 |CG| r 2 − (r − 1)2 2r 2 = = < < = √ . |OG| |OG| 2 r r r Now it is clear that

√ 2 |BC| = √ , for r > 4. δ(r) ≤ |BC| ≤ |BF | r

Thus limr→∞ δ(r) = 0. Problem 3.49. Consider a curve C of length l that divides the surface of the unit sphere into two parts of equal areas. Show that l ≥ 2π . Solution 3.49. Let C be the symmetric (opposite, or antipodal) of C with respect to the center O of the sphere. If C ∩ C = ∅, then int C ∩ int C = ∅ and thus the area of the sphere is strictly greater than the sum of areas of int C and int C . By hypothesis, the latter is just the area of the sphere, which is a contradiction. Therefore, C ∩ C contains at least one point, say P . The point P , the symmetric of P with respect to the symmetry center of C ∩C , should also belong to C ∩C , since the later is symmetric. Finally, note that the shortest curve on the sphere that joins two opposite points is an arc of a great circle, and since C is made of two different arcs joining P and P , it follows that l ≥ 2π. Comments 69 More generally, let K be a body with a symmetry center O and set r = minP ∈∂K |OP |, where ∂K denotes the boundary surface of K. If C ⊂ ∂K is a curve that separates two regions on ∂K of the same area, then the length l of C satisfies l ≥ 2π r.

7.2 Combinatorial Geometry

167

Problem 3.50. Let K be a planar closed curve of length 2π . Prove that K can be inscribed in a rectangle of area 4. Solution 3.50. Let bK (θ ) be the width of the figure K in the direction θ , identified with a vector θ ∈ S 1 . If R(K, θ ) denotes the area of the rectangle circumscribing K and having one side parallel to θ , then π . R(K, θ ) = bK (θ )bK θ + 2 Thus

2π 1 1 π 1/2 bK (θ ) 2 bK θ + dθ, θ 2π 0 2 with equality holding iff bK (θ )bK θ + π2 is constant. From the Cauchy–Schwartz inequality, one derives 1

min R(K, θ ) 2 ≤

1 min R(K, θ ) ≤ θ 4π 2

2

2π

bK (θ ) dθ

,

0

the equality holding iff K is a curve of constant width. We are able now to apply the following theorem of Cauchy, which expresses the length L(K) of the curve K in terms of the integral of its width with respect to all directions: 1 2π bK (θ ) dθ. L(K) = 2 0 This implies that 1 L(K)2 ≤ 4, π2 where the second inequality uses the hypothesis L(K) ≤ 2π . min R(K, θ ) ≤ θ

Comments 70 We can consider the parallelograms with a given acute angle φ instead of rectangles. The same argument shows that min Rφ (K, θ ) ≤ θ

cos φ L(K)2 . π2

See also: •

E. Lutwak: On isoperimetric inequalities related to a problem of Moser, Amer. Math. Monthly 86 (1979), 476–477.

Problem 3.51. 1. Consider a family of plane convex sets with area a, perimeter p, and diameter d. If the family covers the area A, then there exists a subfamily with a pairwise disjoint interiors that covers at least area λA, where λ = a+pd+πd 2. 2. Assume that any two members of the family have nonempty intersection. Prove that there exists then a subfamily with pairwise disjoint interiors that covers area at least μA, where μ = πda 2 .

168

7 Geometry Solutions

Solution 3.51. 1. If K is a member of the family and K(d) the set of points at distance less than d from K, then it is known that area(K(d)) = a + pd + π d 2 . Let {K1 , K2 , . . . , Kn } be a maximal subfamily with disjoint interiors. Then every member of the family intersects some Ki , and thus it is contained in some Ki (d). Thus the family is contained within ni=1 Ki (d). Thus the area covered by all members of the family is at most n · area(Ki (d)) ≤ A, while the subfamily above covers na ≥ λA. 2. The diameter of the union of all sets is at most 2d, by the assumption. Its area is therefore at most πd 2 , by the well-known isoperimetric inequality. Thus λA ≤ a, and one single set will cover at least λA. Comments 71 It is still unknown whether the second claim holds without any additional assumption on the sets. Problem 3.52. Let C be a regular polygon with k sides. Prove that for every n there exists a planar set S(n) ⊂ R2 such that any subset consisting of n points of S(n) can be covered by C, but S(n) itself cannot be included in C. Solution 3.52. Let r be the radius of the circle inscribed in C. For n ∈ Z+ , let S(n) be the circle of radius r sec(π/2kn).

1. C cannot cover S(n). In fact, there is no disk of radius greater than r that can be included in C. Here is a proof for this somewhat obvious assertion. For a fixed radius R, let X(R) be the set of centers of circles of radius R included in C. Then X(R) is convex and invariant of rotations centered at the center O of C, of angle 2π/k. Thus X(R) is either empty or contains O. Thus R ≤ r. 2. Let P1 , . . . , Pn ⊂ S(n) be arbitrary points of S(n). We show that C can cover π from O. We place C P1 , . . . , Pn . We fix a point Q in C at a distance of r sec 2kn such that their centers coincide and we rotate C around its center O such that Q runs through S(n). While Q travels along the circle, some of the points Pi will be in the interior of C and the others outside. However, we see that Pi ∈ / C if and only if Q ∈ nj=1 Aj , where each Aj is an π . More precisely, the arc Aj is arc of a circle from outside C, which is of length kn determined by the inequalities πj π πj ≤ P + . i OQ ≤ 2 k k kn n Therefore j =1 Aj is a set of arcs of total length smaller than nkπ/kn · π and thus S − nj=1 Aj has length π, and thus it is nonempty. Now rotate C until Q lies inside S − nj=1 Aj . Then all the Pj belong to C, as claimed. 2

7.2 Combinatorial Geometry

169

Problem 3.53. Let M be a convex polygon and let S1 , . . . , Sn be pairwise disjoint disks situated in the interior of M. Does there exist a partition M = D1 ∪ · · · ∪ Dn such that Di are convex disjoint polygons, each of which contains precisely one disk? Solution 3.53. Let Oi , ri be the center and the radius of the disk Si . For a point X in the plane, let us define hi (X) = XOi2 − ri2 . We consider Dk = {x ∈ M; hk (X) ≤ hi (X), for all i = 2, 3, . . . , n}. We will prove that the Di are convex and Si ⊂ Di . Fix k, for instance k = 1. Define Hi = {X ∈ R2 ; hk (X) ≤ hi (X)}. Then we claim that Hi are half-planes. Let i = 2 for simplicity. Choose the point P ∈ |O1 O2 | such that |P O1 |2 − r12 = |P O2 |2 − r22 . Take for instance |P O1 | = |O1 O2 |2 −(r12 −r22 ) . If l is the line perpendicular to O1 O2 at P and X ∈ l, then we have 2|O1 O2 | |XO1 |2 − r12 = |P O1 |2 + |P X|2 − r12 = |P O2 |2 + |P X|2 − r22 = |XO2 |2 − r22 . If H˜ 2 is the half-plane that contains O1 , then |XO1 |2 − r12 ≤ |XO2 |2 − r22 , for all X ∈ H˜ 2 , |XO1 |2 − r12 ≥ |XO2 |2 − r22 for all X ∈ / H˜ 2 . Therefore, H2 is the half-plane

H˜ 2 . Consequently,

Dk = M

n 1

Hn

i=1,i=k

is a convex set as an intersection of convex sets, and Sk ⊂ Dk . It is clear that Di = M. Problem 3.54. Consider an inscribable n-gon partitioned by means of n − 2 nonintersecting diagonals into n − 2 triangles. Prove that the sum of the radii of the circles inscribed in these triangles does not depend on the particular partition. Solution 3.54. If we have a triangle inscribed in a circle of center O and radius R, of inradius r, then the distances from O to the sides satisfy the well-known formula d ∗ (O, side) = R + r. cyclic

Here d ∗ is the signed distance, which is positive on one half-plane determined by the side and negative on the other half-plane. We write the corresponding formulas for each triangle of the partition and sum up all the terms. Let d1 , . . . , dn be the distances from O to the sides of the n-gon. When looking at the signed distances from O to a diagonal, we observe that each signed distance is considered twice: once with the plus sign and once with the opposite sign. When these terms are summed, they will cancel. We then obtain d1 + · · · + dn = (n − 2)R + r1 + · · · + rn , and thus the sum of inradii r1 + r2 + · · · + rn does not depend on the partition.

170

7 Geometry Solutions

Problem 3.55. Prove that in an ellipse having semiaxes of lengths a and b and total length L, we have L > π(a + b). Solution 3.55. We cut the ellipse along the axes into four congruent pieces. We can further rearrange the four pieces by adding a square of side length b − a as in the figure below:

We want to use the well-known Bonnesen’s isoperimetric inequality, which states that L2 ≥ 4πA for any planar convex domain of perimeter L and area A. The area of the new domain is A + (b − a)2 , where A was the area of the ellipse. Moreover, one knows that the area of the ellipse is A = π ab. Applying this inequality to the domain pictured in the figure above, we obtain L2 > 4π 2 ab + 4π(b − a)2 > π 2 (a + b)2 , whence the claim. Problem 3.56. Let F be a convex planar domain and F denote its image by a homothety of ratio − 21 . Is it true that one can translate F in order for it to be contained in F ? Can the constant 21 be improved? Generalize to n dimensions. Solution 3.56. Let ABC ⊂ F be a triangle of maximal area inscribed in F . Let O denote the geometric centroid of ABC. We use the homothety of center O and ratio − 21 that transforms F into F and ABC into A B C . Since O is the centroid of ABC, we find that A is the midpoint of the segment |BC|, B is the midpoint of |CA|, and C is the midpoint of |AB|.

7.2 Combinatorial Geometry

171

We will show that F ⊂ ABC. Assume the contrary, namely that there exists a point X of F that does not belong to the triangle ABC. By symmetry, it suffices to consider that X belongs to the half-plane determined by the line BC, which does not contain A. Since BC and B C are parallel, it follows that the area of XB C is strictly greater than the area of A B C . Let us consider the inverse image Y of X by the homothety of center O and ratio − 21 . Then Y belongs to F and the area of Y BC is strictly greater than the area of ABC, which contradicts our assumptions. This proves that F ⊂ ABC ⊂ F . The constant 21 is sharp in the case that F is a triangle. The same proof shows that the image F of a convex domain F in Rn by the homothety of ratio − n1 can be translated in order to be contained in F . The constant 1 n is sharp, with equality for the simplex. Problem 3.57. A classical theorem, due to Cauchy, states that a strictly convex polyhedron in R3 whose faces are rigid, must be globally rigid. Here, rigidity means continuous rigidity, in the sense that any continuous deformation of the polyhedron in R3 that keeps the lengths of edges fixed is the restriction of a deformation of rigid Euclidean motions of three-space. Prove that a 3-dimensional cube immersed in Rn remains rigid for all n > 3. Solution 3.57. Three sides are common to a vertex and determine a three-dimensional space R3 ⊂ Rn . The 3-dimensional spaces determined by two adjacent vertices have two sides in common, and therefore they coincide. Thus, the entire cube is immersed in a 3-dimensional subspace of Rn , and therefore it remains rigid. The same argument shows that a rigid polyhedron with the property that exactly three sides meet at any vertex remains rigid after immersion in Rn , for all n > 3. Problem 3.58. Consider finitely many great circles on a sphere such that not all of them pass through the same point. Show that there exists a point situated on exactly two circles. Deduce that if we have a set of n points in the plane, not all of them lying on the same line, then there must exist one line passing through precisely two points of the given set. Solution 3.58. Assume that the contrary holds. Consider the partition of the sphere into (spherical) polygons, induced by the circles, and identify it with a polyhedron. Each vertex will have degree at least 6, since at least three great circles pass through each point. If V is the number of vertices, E the number of edges, and F the number of faces of the spherical polyhedron, then Euler’s formula states that V − E + F = 2. Moreover, by counting the edges in each face and summing over the faces, we obtain 2E = 3F3 + 4F4 + · · · ≥ 3F, where Fk is the number of faces with k sides. Finally, counting the edges adjacent to each vertex and summing over the vertices, we get

172

7 Geometry Solutions

2E = 6V6 + 7V7 + · · · ≥ 6V , where Vk is the number of vertices of degree k, because Vk = 0 when k ≤ 5. Thus 6(E + 2) = 6(V + F ) ≤ 2E + 4E = 6E, which is absurd. Therefore, there exists a vertex of valence smaller than 6 in which exactly two great circles intersect. The second claim can be reformulated using projective duality: if we have n lines in the projective plane, not all passing through the same point, then there exists a point lying on exactly two of the lines. This is equivalent to the problem considered above. Comments 72 The problem was posed by Sylvester in 1839, and it was not settled until 1930, when T. Gallai gave a solution. The solution above is due to N. Steenrod. Moreover, it can be proved that the number of points where exactly two circles intersect is at least 3n/7; see for instance: • •

H.S.M. Coxeter: The classification of zonohedra by means of projective diagrams, J. Math. Pures Appl. (9) 41 (1962), 137–156. G.D. Chakerian, Sylvester problem on collinear points and a relative, Amer. Math. Monthly 77 (1970), 164–167.

Problem 3.59. Given a finite set of points in the plane labeled with +1 or −1, and not all of them collinear, show that there exists a line determined by two points in the set such that all points of the set lying on that line are of the same sign. Solution 3.59. We use projective duality as above and reformulate the question as follows: given n great circles on a sphere, to each one being assigned either +1 or −1, and not all passing through the same point, show that there exists a point lying on at least two circles such that all circles containing that point have the same sign. The circles determine a spherical polyhedron whose edges are labeled by +1 and −1. Assume that the result is not true. Then the edges around each vertex must have both the labels +1 and −1. By symmetry, the number of sign changes in the labels of the edges that one meets when traveling-cyclically around each vertex is at least 4. Denote by N the sum of all such sign changes over all vertices. Then N ≥ 4V . On the other hand, by counting the sign changes as we travel around the faces and summing over all faces, we should again obtain N . Further, the number of sign changes in a k-sided face is even and at most k, so that N ≤ 2F3 + 4F4 + 4F5 + 6F6 + 6F7 + · · · . Moreover, using Euler’s formula and 2E = 3F3 + 4F4 + · · · we obtain 4V − 8 = 4E − 4F = 2(3F3 + 4F4 + 5F5 + · · · ) − 4(F3 + F4 + · · · ) = 2F3 + 4F4 + 6F5 + · · · ≥ N ≥ 4V , which is a contradiction. Thus our claim follows.

7.2 Combinatorial Geometry

173

Comments 73 The result concerning the sign changes is valid more generally when some edges are allowed to be unlabeled, and as such it is known as Cauchy’s lemma. This is an essential ingredient in proving the famous rigidity theorem of Cauchy, which states that a convex polyhedron in R3 whose faces are rigid is globally rigid. Problem 3.60. If Q is a given rectangle and ε > 0, then Q can be covered by the union of a finite collection S of rectangles with sides parallel to those of Q in such a way that the union of every nonoverlapping subcollection of S has area less than ε. Solution 3.60. Observe first that every rectangle Q has a subset H that is the union of a finite collection of rectangles with parallel sides such that: 1. any two members of overlap; 2. area(H ) > δ area(Q), where δ = exp(−1/ε); 3. area(A) < 2ε area(H ), for any A ∈ . If Q = [0, a] × [0, b], we consider rectangles Ai that are parallel and have one vertex at (0, 0) and the other vertex on the curve of equation xy = δ area(Q). Thus each rectangle has area δ. The curved triangle determined by the inequalities xy ≤ δ area(Q), x, y ≥ 0, will be denoted by T . Moreover, we consider a collection of such rectangles such that their union H satisfies area(H ) > 1+ε ε area(T ) = δ area(Q). Further, area(A) = δ area(Q) < ε area(T ). Set Q0 = Q, H0 = H , 0 = and assume that area(Q) = 1, for simplicity. Then Q0 − H0 is a finite union of parallel rectangles Qi,0 and area(Q0 − H0 ) < (1 − δ). Choose in each rectangle Qi a set of type Hi,0 as above. The union of all such Hi,0 and H0 will be denoted by H1 , and the collection of rectangles will be denoted by 1 . Then Q − H1 is also a finite union of parallel rectangles Qi,1 and area(Q − H1 ) < (1 − δ)2 area(Q). We continue this process by defining the sets Hj and the collections j of rectangles such that Q − Hj is a finite union of rectangles Qi,j whose complement has total area area(Q − Hj ) < (1 − δ)j area(Q). We stop this process at the stage n, where n is large enough that (1 − δ)n area(Q) < ε/2. Consider then the union of the collections n obtained so far with the set of rectangles in which Q − Hn decomposes. If we consider a subfamily of disjoint rectangles from , then there is at most one element Ai from any collection i,j covering Hi,j , because all elements in i,j overlap each other. Thus the total area covered by these Ai is at most ε ε area(Hi,j ) < . 2 2 i,j

Since the rectangles from Q − Hn contribute at most 2ε , we are done.

174

7 Geometry Solutions

Problem 3.61. Prove that the 3-dimensional ball cannot be partitioned into three sets of strictly smaller diameter. Solution 3.61. Assume that we can divide the unit ball into three pieces of smaller diameter and let X1 , X2 , X3 be the partition that it induces on the boundary sphere. Thus the diameter of each piece Xi is smaller than 1 − 10h, for some h > 0. We choose a very thin triangulation of the sphere with triangles that have edges smaller than h. Denote by Nj the union of those spherical triangles of the triangulation above that intersect the set Xj nontrivially. It follows that N1 , N2 , N3 are closed subsets of the sphere that are domains, i.e., they are closures of open subsets. In particular, the boundary of each domain Nj is the union of several piecewise linear circles. Moreover, the union N1 ∪ N2 ∪ N3 covers the sphere. Since each triangle has a small size, the diameter of Nj is at most 1 − 8h < 1. In particular, opposite points cannot belong simultaneously to the same set Nj . Let us denote by x ∗ the point of the sphere opposite the point x. We know that the set N1 and its symmetric N1∗ (which is the set of those points x ∗ for which x ∈ N1 ) should be disjoint; otherwise, there would exist in N1 a pair of points at distance 1. Consider the circles determined by the boundary circles of both N1 and N1∗ . The circles associated to Nj are disjoint and distinct from the circles associated to N1∗ , because N1 ∩ N1∗ = ∅. Thus, there is an even number of such circles. Observe that k disjoint circles on the sphere divide the surface of the sphere into k + 1 connected complementary regions. Thus, the collection of boundary circles associated to N1 and N1∗ will split the surface of the sphere into an odd number of connected regions, which we denote by R1 , R2 , . . . , R2m+1 . Consider the symmetric Rj∗ of the region Rj . There are two possibilities: either ∗ Rj is another connected region Rk that is disjoint from Rj , or else Rj∗ = Rj , meaning that Rj is symmetric. The second case happens, for instance, when the region Rj is bounded by two circles, one coming from N1 and the other from N1∗ . If there were no symmetric regions Rj , then the total number of regions would be even, which would contradict our previous claim. Thus there must be at least one connected symmetric region Rj . We consider a point x ∈ Rj and its symmetric x ∗ , which must also belong to Rj , since the region is symmetric. If x ∈ N1 , then all points of the sphere that are connected to x by a path not hitting a boundary circle of N1 must also belong to N1 . Thus, the connected component Rj will be contained in N1 , and so N1 will contain pairs of symmetric points, contradiction. This implies that x ∈ N1 and hence x ∈ N2 ∪ N3 . Let us assume that x ∈ N2 . Then its symmetric x ∗ is not in N2 , since N2 has diameter smaller than 1 and thus x ∗ ∈ N3 . Further, Rj is connected and thus there exists a path γ within Rj that connects x to x ∗ . One knows that N2 and N3 are domains covering the path γ . Moreover, there exists a point z ∈ γ that belongs to both N2 and N3 . In fact, consider the point z ∈ γ closest to x ∗ and that belongs to N2 . There exists such a point since N2 is closed. If z = x ∗ , then both x and x ∗ belong to N2 , contradiction. If z = x ∗ , then the points on the right of z should belong to N3 and thus z belongs to the closure of N3 and thus to N3 .

7.3 Geometric Inequalities

175

Now z ∈ Rj and thus its symmetric z∗ also belongs to Rj . This means that ∈ N1 . But z∗ can belong neither to N2 , since N2 does not contain opposite points, nor to N3 , for the same reason. This contradiction proves the claim. z, z∗

Comments 74 More generally, it is known that the unit n-ball (or equivalently, the (n − 1)-sphere in Rn ) cannot be partitioned into n pieces of strictly smaller diameter. This is related to Borsuk’s problem discussed in comments to Problem 3.38. The result was proved by Lusternik and Shnirelman in 1930 and, independently, by Borsuk in 1932, but the higher-dimensional case n ≥ 4 requires deeper methods from topology, and this opened a new chapter in algebraic topology. The interested reader might consult the survey of Steinlein, which presents both historical remarks and a large number of references. • • •

•

K. Borsuk: Über die Zerlegung einer n-dimensionalen Vollkugel in n-Mengen, Verh. International Math. Kongress Zürich 1932, 192. L.A. Lusternik and L.G. Shnirelman: Topological Methods for Variational Problems, Moscow, 1930. H. Steinlein: Borsuk’s antipodal theorem and its generalizations and applications: a survey, Topological Methods in Nonlinear Analysis, 166–235, Sém. Math. Sup., 95, Presses Univ. Montreal, Montreal, 1985. H. Steinlein: Spheres and symmetry: Borsuk’s antipodal theorem. Topol. Methods Nonlinear Anal. 1 (1993), 1, 15–33.

7.3 Geometric Inequalities Problem 3.62. If a, b, c, r, R are the usual notations in the triangle, show that 1 1 1 1 2 1 ≤ ≤ 2. ≤ 2rR 3 a a2 4r Solution 3.62. We derive from ( a1 − b1 )2 ≥ 0 that 1 1 1 2 1 ≤ . ≤ bc 3 a a2 Moreover, the usual Cauchy–Schwarz inequality shows that (p − a)(p − b) ≤ and its analogues. Then we have 1 2p a+b+c 1 1 1 1 1 2 1 ≤ = = = + + ≤ 2rR abc abc ab bc ca 3 a a2 1 1 p 1 ≤ = = 2. 4 (p − b)(p − c) 4(p − a)(p − b)(p − c) 4r Problem 3.63. If a, b, c are the sides of a triangle, then prove that b2 +c2

ma ka , where ma , wa , ka

2bc ≤ and altitude issued from A.

(b+c)2 4bc

≤

ma wa

c2 4,

and

denotes the respective lengths of the median, bisector,

176

7 Geometry Solutions

Solution 3.63. We have the following formulas computing ma , wa , and ka : √ " bc 1 2 2 2 2 b + c − a , wa = (b + c)2 − a 2 , ma = 2 b+c " bc ka = 2 2 b2 + c2 − a 2 . b + c2 From the triangle inequality, we have a 2 > (b − c)2 , which implies furthermore that (b + c)2 − a 2 < 4bc

(b − c)2 (b − c)2 , ≤ 4bc (b + c)2 − a 2

with equality only if b = c. Adding 1 in both members yields / 2b2 + 2c2 − a 2 b+c 2b2 + 2c2 − a 2 (b + c)2 ≤ ; hence . ≤ √ 4bc (b + c)2 − a 2 (b + c)2 − a 2 2 bc This leads to (b + c)2 b+c ≤ √ 4bc 2 bc

/

2(b2 + c2 ) − a 2 ma − a2 = . 2 (b + c) wa

For the second inequality, it suffices to observe that ma ma b2 + c2 ≥ = . ha ka 2bc Notice that equality in the first case implies b = c, while for the second case either b = c or a 2 = b2 + c2 . Problem 3.64. If S(x, y, z) is the area of a triangle with sides x, y, z, prove that S(a, b, c) + S(a , b , c ) ≤ S(a + a , b + b , c + c ). Solution 3.64. We make the following change of variables: s = (a + b + c)/2, t = s − a, u = s − b, and v = s − c. Using Heron’s formula, the inequality becomes √ √ 4 4 stuv + s t u v ≤ 4 (s + s )(t + t )(u + u )(v + v ). We prove first that for positive x, x , y, y , we have √ xy + x y ≤ (x + x )(y + y ). In fact, this follows from xy + x y + 2 x y · y x ≤ xy + x y + xy + yx , and equality holds when the geometric means of√x y and y x are √ equal. By applying the last inequality to x = st and y = uv twice, we obtain the claim. Moreover, the equality is attained only when t/t = u/u = v/v = s/s , which amounts to a/a = b/b = c/c .

7.3 Geometric Inequalities

177

Problem 3.65. It is known that in any triangle we have the inequalities √ √ 3 3r ≤ p ≤ 2R + (3 3 − 4)r, where p denotes the semiperimeter. Prove that in an obtuse triangle, we have √ (3 + 2 2)r < p < 2R + r. Solution 3.65. We have

cos A−B a+b 2 . = c sin C2 √ √ Since C > π2 , we have sin C2 > 22 and thus c 2 > a + b. Therefore √ √ 2p 2 > (a + b)( 2 + 1) ⇒ 8p 2 √ √ > (3 + 2 2)(a + b)2 > 4(3 + 2 2)ab √ > 8(3 + 2 2)S, where S is the area. This implies that √ √ 3+2 2 S = 3 + 2 2 r. p> p Finally, recall that r = (p − c) tan 2R + r.

C 2

> (p − c) and so p < c + r = 2R sin C + r

2, we have |an | > 2 and consequently an+1 > an for all n ≥ 2. Since the sequence is increasing, it is either convergent or tends to ∞. If we have a finite limit an → λ, then it satisfies λ = λ2 − 2 and thus λ ∈ {−1, 2}, contradicting our assumptions. Therefore, we need |a| ≤ 2 for the convergence. Let α ∈ − π2 , π2 be such that 2 cos α = a. By induction on n, we obtain an = 2 cos 2n α. / Q. Observe that the Assume that α is not commensurable with π, that is, πα ∈ function ϕ(x) = 2x α is continuous. Thus, the map ϕ : R/π Z → R/π Z sends the dense subset {nα (mod π), n ∈ Z} into a dense subset, which is {2n α (mod π ), n ∈ Z}. Therefore, the image under cos, namely {cos 2n α; n ∈ Z}, is dense in [−1, 1], and thus the sequence (an ) cannot be convergent. If α/π = p/q, where p, q ∈ Z, then the sequence (an ) is periodic for n large enough. Thus, in order for the sequence to be convergent it is necessary and sufficient that it be constant for n large enough. This condition reads

204

8 Analysis Solutions

cos 2n α = cos 2n+1 α . Thus 2n α = ±2n+1 α + kπ,

where k ∈ Z,

) ( m π, where m, n ∈ Z . 2n 3 One sees that an is convergent and constant for n ≥ m.

and so

α∈

Problem 4.33. Let 0 < a < 1 and I = (0, a). Find all functions f : I → R satisfying at least one of the conditions below: 1. f is continuous and f (xy) = xf (y) + yf (x). 2. f (xy) = xf (x) + yf (y). Solution 4.33. 1. Let g(x) = f (x) x . Then g(exp(x)) is continuous and additive on I . This implies that g(exp(x)) is linear and hence g(x) = C log x, which implies f (x) = Cx log x for some constant C. 2. Give y the values x, x 2 , x 3 , and x 4 . We find therefore that f x 2 = 2xf (x), f x 3 = xf (x) + x 2 f x 2 = x + 2x 3 f (x), f x 4 = xf (x) + x 3 f x 3 = x + x 4 + 2x 6 f (x), f x 4 = x 2 f (x) + x 2 f x 2 = 4x 3 f (x). From the last two lines, one derives that f (x) = 0 for all but those points x for which x + x 4 + 2x 6 = 4x 3 . This means that f vanishes for all but finitely many (actually 6) values of x. If t ∈ I is such that f (t) = 0, then the first identity shows that f t 2 = 0. n By induction on n we obtain f (t 2 ) = 0. This means that f does not vanish for infinitely many values of the argument, which contradicts our former result. Thus f is identically zero. Problem 4.34. If a, b, c, d ∈ C, ac = 0, prove that −1 + max(|ac|, |ad + bc|, |bd|) ≥ max(|a|, |b|) max(|c|, |d|) 2 Solution 4.34. Set ab−1 = r, dc−1 = s, −1+2 inequality from the statement is equivalent to

√ 5

√ 5

.

= k, k 2 = 1 − k, 0 < k < 1. The

μ = max(1, |r + s|, |rs|) ≥ k max(1, |r|) max(1, |s|) = ν. (i) If |r| ≥ 1, |s| ≥ 1, then μ ≥ |rs| > k|s||r| = ν,

8 Analysis Solutions

205

proving our claim. (ii) If |r| ≤ 1, |s| ≥ 1, then our inequality is equivalent to max(1, |r + s|, |rs|) ≥ k|s|. Moreover, if k|s| ≤ 1 or |r + s| ≥ k|s|, then this is obviously true. Further, let us assume that k|s| > 1 and |r +s| < k|s|. We have then |r|+|r +s| ≥ |s| and consequently, |r| ≥ |s| − |s + r| ≥ |s| − k|s| = (1 − k)|s| = k 2 |s|. Furthermore, |rs| ≥ k 2 |s|2 > k|s| > 1; thus max(1, |r + s|, |rs|) ≥ k|s| holds. (iii) If |r| ≥ 1, |s| ≤ 1, then the inequality is proved as in case (ii), by using the obvious symmetry. Problem 4.35. Let ∞ i=1 xi be a convergent series with decreasing terms x1 ≥ x2 ≥ · · · ≥ xn ≥ · · · > 0 and let P be the set of numbers which can be written in the form i∈J xi for some subset J ∈ Z+ . Prove that P is an interval if and only if xn ≤

∞

xi for every n ∈ Z+ .

i=n+1

Solution 4.35. that xp > ∞ i=p+1 xi and consider 1. Suppose that there exists p such α such that i>p xi < α < xp . Set S(J ) = x . We claim that there does i∈J i such that S(J ) = α. In fact, if J ∩ {1, 2, . . . , p} = ∅, then not exist any J ⊂ Z + x ≥ x > α, because x is decreasing. On the other hand, if J ∩{1, 2, . . . , p} = i p k i∈J ∅, then i∈J xi ≤ i>p xi < α. Finally, P contains numbers smaller than α, for instance i>p xi , as well as numbers bigger than α, such as xp . Thus P is not an interval. 2. Assume now that the hypothesis of the problem ∞ is satisfied and choose any element y from the interval (0, S], where S = i=1 xi . We will show that there exists L such that S(L) = y. Let n1 be the smallest index such that xn1 < y. There exists such a one because limk→∞ xk = 0. By induction, we define nk to be the smallest integer with the property that xn1 + · · · + xnk < y. Let L = {n1 , . . . , nk , . . . }. It is clear that S(L) ≤ y. If p ∈ Z+ \ L, then choose the smallest k with the property that nk > p. Since p does not belong to L, we have xn1 + · · · + xnk−1 + xp ≥ y. This implies that

$∞ % xi + xp ≥ y. i∈L

206

8 Analysis Solutions

(i) Assume that the set Z+ − L is finite. This set is empty only when y = S, which obviously belongs to P . Now take p to be the maximal element of Z+ − L. This implies that all elements bigger than p already belong to L and hence L = {n1 , . . . , nk−1 } ∪ {p + 1, p + 2, . . . }. Then

xi =

i∈L

k−1 i=1

xn1 +

∞

xj ≥

j =p+1

k−1

xni + xp ≥ y,

i=1

from which we obtain that i∈L xi = y. (ii) If Z+ − L is infinite, then for any ε > 0, one can choose some pε ∈ Z+ − L such that xpε < ε. The inequality above for pε implies that xi ≥ y. + i∈L

This is true for any positive ε, and thus

i∈L xi

= y.

Problem 4.36. Does there exist a continuous function f : (0, ∞) → R such that f (x) = 0 if and only if f (2x) = 0? What if we require only that f be continuous at infinitely many points? Solution 4.36. 1. There exists some real number α such that f (α) = 0, and thus f (2α) = 0. Let γ ∈ [α, 2α] be given by γ = sup{x ∈ [α, 2α] such that f (x) = 0}. Since f is continuous, we find that f (γ ) = 0, and thus γ < 2α. Moreover, if xn ∈ R is a sequence such that limn xn = γ , then f (xn ) = 0 for n large enough. In fact, otherwise there exists a subsequence f (xnk ) = 0, which implies that f (2xnk ) = 0, and thus limk f (2xnk ) = 0 = f (2γ ). Finally, recall that γ is the supremum of those x ∈ [α, 2α] for which f (x) = 0 and γ < 2α. Thus for n large such that γ + n1 < 2α, there exists xn with γ < xn < γ + n1 for which f (xn ) = 0. This contradicts the previous claim. Thus there does not exist a continuous f as in the statement. 2. The function ' 0, for 22k < x ≤ 22k+1 , where k ∈ Z, f (x) = 1, for 22k−1 < x < 22k , where k ∈ Z, satisfies the claim and has a countable number of discontinuities. Problem 4.37. Find the smallest number a such that for every real polynomial f (x) of degree two with the property that |f (x)| ≤ 1 for all x ∈ [0, 1], we have |f (1)| ≤ a. Find the analogous number b such that |f (0)| ≤ b. Solution 4.37. Let f (x) = ux 2 + vx + w. We have |f (1)| = |2u + v| = |3f (1) − 4f (1/2) + f (0)| ≤ 3|f (1)| + 4|f (1/2)| + |f (0)| ≤ 3 + 4 + 1 = 8.

8 Analysis Solutions

207

Further, a = 8 because we have equality above for f (x) = 8x 2 − 8x + 1. For the second case, we observe that |f (0)| = |v| = |4f (1/2) − 3f (0) − f (1)| ≤ 8 and thus b = 8 by taking f (x) = 8x 2 − 8x + 1. Problem 4.38. Let f : R → R be a function for which there exists some constant M > 0 satisfying |f (x + y) − f (x) − f (y)| ≤ M, for all x, y ∈ R. Prove that there exists a unique additive function g : R → R such that |f (x) − g(x)| ≤ M, for all x ∈ R. Moreover, if f is continuous, then g is linear. Solution 4.38. Let us show first the existence of the upper limit ϕ(x) = lim supn→∞ f (nx) n . In fact, using the hypothesis, one finds that f (nx) ≤ n(f (x) + M), and hence the sequence f (nx) n is bounded from above, and thus ϕ(x) exists. Furthermore, f (n(x + y)) f (nx) f (ny) M ≤ − , − n n n n which yields ϕ(x) + ϕ(y) = ϕ(x + y); thus ϕ is additive. Next, observe that f (nx) ≤ n(f (x) + M) implies that nx n − f (x) ≤ M, and we can take g(x) = ϕ(x). If f is continuous, then it is easy to see that the lower limit, lim inf n→∞ f (nx) n , coincides with ϕ(x), and moreover, ϕ(x) is continuous. An additive continuous function is therefore linear. In fact, we have f (nx) = nf (x) for n ∈ Z, which yields f (rx) = rf (x) for any r ∈ Q. Since f is continuous and every real number r can be approximated by a sequence of rational numbers, we find that f (rx) = rf (x) for any r ∈ R, and hence the claim. Let us prove the uniqueness of such additive functions. Assume that there exists another additive function h satisfying |f (x) − h(x)| ≤ M, for all x ∈ R. This implies that |g(x) − h(x)| ≤ M, for all x ∈ R. However, an additive function is either zero or unbounded. Therefore, the additive function g − h must vanish.

208

8 Analysis Solutions

Comments 88 Functions f satisfying the condition from the statement are called quasimorphisms. Notice that there exist additive functions ϕ : R → R that are not linear (not continuous of course). Examples can be constructed by choosing a basis of R as a vector space over Q and defining arbitrarily the values of the function on each vector of that basis. Problem 4.39. Show that if f is differentiable and if lim f (t) + f (t) = 1, t→∞

then lim f (t) = 1.

t→∞

Solution 4.39. Let us show first that if limt→∞ (f (t) + αf (t)) = 0, where a = (α) > 0, then limt→∞ f (t) = 0. Let ε > 0 and c < ∞ be such that |f (t) + αf (t)| ≤ aε whenever t ≥ c. Then d αt αt = e (f (t) + αf (t)) ≤ aεeat f (t) e dt for t ≥ c. Therefore, using the mean value theorem we have |eαt f (t) − eac f (c)| ≤ ε eαt − eac , for t > c. This yields

|f (t)| ≤ ea(c−t) · |f (c)| + ε1 − ea(c−t) .

In particular, for sufficiently large t, we have |f (t)| ≤ 2ε, as claimed. Now the statement follows by applying this result to f − 1. Comments 89 Let P (D) be a polynomial in the derivation operator D = write P (D) = (D − λ1 )(D − λ2 ) · · · (D − λn ).

d dt .

Let us

The above result can be generalized as follows. Assume that λi < 0, where denotes the real part; if the function f satisfies limt→∞ P (D)f (t) = 0, then limt→∞ f (t) = 0. This is proved by recurrence on the degree of P . The recurrence step follows from the argument given above. In particular, the claim holds for P (D) = D2 + D + 1. The condition λi < 0 is necessary. In fact, we can consider f (t) = eλi t , which satisfies limt→∞ P (D)f → 0. Also, one notices that P (D) = D0 + D1 + ... + Dn does not fulfill this condition for n ≥ 3. Problem 4.40. Let c be a real number and let f : R → R be a smooth function of class C 3 such that limx→∞ f (x) = c and limx→∞ f (x) = 0. Show that limn→∞ f (x) = limx→∞ f (x) = 0.

8 Analysis Solutions

209

Solution 4.40. According to Taylor’s formula, there exist ξx , ηx ∈ (0, 1) such that 1 f (x + 1) = f (x) + f (x) + f (x) + 2 1 f (x − 1) = f (x) − f (x) + f (x) − 2 Suitable linear combinations of these yield

1 f (x + ξx ), 6 1 f (x + ηx ). 6

1 1 f (x) = f (x + 1) − 2f (x) + f (x − 1) − f (x + ξx ) + f (x − ηx ), 6 6 1 1 2f (x) = f (x + 1) − f (x − 1) − f (x + ξx ) − f (x − ηx ). 6 6 Since x + ξx → ∞ and x − ηx → ∞ as x → ∞, by passing to the limit, one obtains 1 1 lim f (x) = c − 2c + c − · 0 + · 0 = 0, 6 6 1 1 1 lim f (x) = c − c − · 0 − · 0 = 0. x→∞ 2 6 6

x→∞

Comments 90 It can be proved that if limx→∞ f (x) exists and if f (n) is bounded, then limx→∞ f (k) (x) = 0 for any 1 ≤ k < n. See for instance: •

J. Littlewood: The converse of Abel’s theorem, Proc. London Math. Soc. (2) 9 (1910–11), 434–448.

Problem 4.41. Prove that the following integral equation has at most one continuous solution on [0, 1] × [0, 1]: x y f (x, y) = 1 + f (u, v) du dv. 0

0

Solution 4.41. If f1 , f2 are two solutions, then their difference g satisfies x y g(x, y) = g(u, v) du dv. 0

0

Since g is continuous, its absolute value is bounded by some constant M on the square [0, 1] × [0, 1]. Thus x y |g(x, y)| ≤ M du dv = Mxy. 0

0 n n

By induction, we show that |g(x, y)| ≤ M x(n!)y 2 . The induction step follows from

|g(x, y)| ≤ 0

x

0

y

Mun v n M(xy)n+1 du dv = . 2 (n!) ((n + 1)!)2

If we fix x, y, then 0 ≤ |g(x, y)| ≤ lim n→

and thus g(x, y) ≡ 0.

Mx n y n = 0, n!n!

210

8 Analysis Solutions

Comments 91 There exists a continuous solution of this equation, given by f (x, y) = 1 + xy +

√ x2y2 x3y3 + + · · · = J0 −4xy = J0 2i xy , 2!2! 3!3!

where J0 is the Bessel function of order zero. Problem 4.42. Find those λ ∈ R for which the functional equation

1

min(x, y)f (y) dy = λf (x)

0

has a solution f that is nonzero and continuous on the interval [0, 1]. Find these solutions. Solution 4.42. The equation can be written in the form

x

λf (x) =

yf (y) dy + x

0

1

f (y) dy. x

If λ = 0, then f is differentiable, and after derivation, we obtain

1

λf (x) = xf (x) − xf (x) +

f (y)dy =

x

1

f (y)dy. x

Thus f is differentiable and λf (x) = −f (x). This implies that f (x) = A cos μx + B sin μx, where μ = λ−1/2 if λ > 0, and f (x) = A cosh νx + B sinh νx, where ν = (−λ)−1/2 if λ < 0. According to the hypothesis, lim f (x) = 0, n→0

and hence A = 0 for any choice of λ. From the above, we also get lim f (x) = 0,

n→0

and thus Bμ cos μ = 0 and Bν cosh ν = 0. The last equation yields B = 0, and so we do not have nonzero solutions. For λ > 0, if B = 0, then cos μ = 0 and thus 4 μ = (2k + 1) π2 . Thus λ = μ−2 (2k+1) 2 π 2 , where k ∈ Z, and π f (x) = B cos(2k + 1) x, B ∈ R. 2 If λ = 0, the same method gives us f (x) = 0.

8 Analysis Solutions

211

Comments 92 We proved that the integral operator T : C 0 (0, 1) → C 0 (0, 1) defined on the space C 0 (0, 1) of continuous functions on (0, 1) and given by the formula 1 min(x, y)f (y)dy (Tf )(x) = 0

has the eigenvalues

4 (2k+1)2 π 2

and the eigenvectors cos(2k + 1) π2 x.

Problem 4.43. Let X be an unbounded subset of the real numbers R. Prove that the set AX = {t ∈ R; tX is dense modulo 1} is dense in R. Solution 4.43. Let {On }n∈Z+ be a countable base of (0, 1), On = ∅. Let p be the canonical projection p : R → R/Z. Define An = {r ∈ R; p(rX) ∩ On = ∅}. This amounts to An =

x∈X−{0}

1 −1 p ({On }), x

and thus An is open. If F is an interval of length f > 0, then there exists x ∈ X such that xf > 1, which implies that p(xF ) = R/Z and thus there exists r ∈ F with p(rx) ∈ On . This proves that An is dense in R. Since AX =

∞ 1

An ,

n=0

we obtain that AX is also dense in R, as the intersection of open dense subsets. Problem 4.44. Consider P (z) = zn +a1 zn−1 +· · ·+an , where ai ∈ C. If |P (z)| = 1 for all z satisfying |z| = 1, then a1 = · · · = an = 0. Solution 4.44. Consider the polynomial Q(z) = 1 + a1 z + · · · + an zn , which can be written as Q(z) = P (z), when |z| = 1. This implies that |Q(z)| = 1 for all z of unit modulus. Moreover, Q(0) = 1, and so Q has a value in the interior of the disk whose modulus is equal to its maximum on the boundary circle. According to the maximum principle for holomorphic functions, Q has to be constant, Q(z) = 1, whence the claim. Comments 93 W. Blaschke has studied the complex analytic (i.e., holomorphic) functions f : D → C on the unit disk D that are continuous on D and have |f (z)| = 1 for all z satisfying |z| = 1. He has shown that such a function has the form f =σ

n & z − bk , 1 − bk z

k=1

where σ is a constant of modulus one and bk ∈ C are such that |bk | < 1.

212

8 Analysis Solutions

Problem 4.45. Let I ⊂ R be an interval and u, v : I → R smooth functions satisfying the equations u (x) + A(x)u(x) = 0, v (x) + B(x)v(x) = 0, where A, B are continuous on I and A(x) ≥ B(x) for all x ∈ I . Assume that v is not identically zero. If α < β are roots of v, then there exists a root of u that lies within the interval (α, β), unless A(x) = B(x), in which case u and v are proportional for α ≤ x ≤ β. Solution 4.45. The roots of v are isolated and we can suppose that v|(α, β) > 0. Assume that u|(α, β), so that after a possible sign change, u|(α, β) > 0. Also, we have v (α) ≥ 0, v(β) ≤ 0, and since v is not identically zero, then v and v do not have common roots; thus v (α) > 0 and v (β) < 0. We consider w(x) = u(x)v (x) − u (x)v(x). We have w (x) = u(x)v (x) − u (x)v(x) = (A(x) − B(x))uv ≥ 0, and therefore w is increasing; so w(α) ≤ w(β) and thus u(α)v (α) ≤ u(β)v (β). Since u(α) ≥ 0, u(β) ≥ 0, v (α) > 0, v (β) < 0, we obtain u(α) = u(β) = 0, and so w(α) = w(β) = 0. Now w is increasing and hence w ≡ 0 and thus w ≡ 0, which u u yields A(x) = B(x). Moreover, for α < x < β we have vw(x) 2 (x) = ( v ) , and hence v is constant. Comments 94 This result is known as Sturm’s comparison theorem in differential equations. Problem 4.46. Let V be a finite-dimensional real vector space and f : V → R a continuous mapping. For any basis B = {b1 , b2 , . . . , bn } of V , consider the set EB = {z1 b1 + · · · + zn bn , where zi ∈ Z}. Show that if f is bounded on EB for any choice of the basis B, then f is bounded on V . Solution 4.46. We solve first the case n = 1. Given any pair of positive numbers 0 < a < b, there exists some r = r(a, b) (depending on a, b) such that for any x > r, one can find m ∈ Z, with ma ≤ x ≤ mb. In fact, the intervals (ma, mb), for integral m, will cover all of R but a compact set. Thus, for any x > r, one can find d an interval (c, d) containing x, and some m ∈ Z such that a < mc < m < b. Assume that f is unbounded. Let us consider an interval I1 = [a1 , b1 ] with the property that f (x) > 1, for x ∈ I1 . Since f is unbounded, there exist x1 > r(a1 , b1 ) arbitrarily large such that f (x1 ) > 4 and m1 ∈ Z such that m1 a1 < x1 < m1 b1 . Moreover, f is continuous and so f (x) > 2 for any x ∈ [c1 , d1 ], for some small interval around x1 , which can be chosen to be contained in (m1 a1 , m1 b1 ). Set I2 =

8 Analysis Solutions

213

{x ∈ I1 ; m1 x ∈ [c1 , d1 ]}. Thus I2 ⊂ I1 is a compact nonempty interval with the property that f (m1 x) > 2 for any x ∈ I2 . We define in this way a nested sequence of compact intervals I1 ⊂ I2 ⊂ I3 ⊂ · · · Ik ⊂ · · · , and a sequence of integers m1 < m2 < m3 < · · · with the property that f (x) > 2k for x ∈ mk Ik . This nested sequence of nontrivial compact intervals must have nontrivial intersection. Let a ∈ ∩∞ k=1 Ik . Then the restriction of f to aZ is unbounded, since for any k, there exists some mk with f (mk a) > 2k , by making use of the fact that a ∈ Ik . Let now deal with the general case. Let zk ∈ V be a sequence for which f (zk ) is f is continuous. unbounded, e.g., f (zk ) > 2k+1 . Then zk is unbounded, because Let B = {b1 , . . . , bn } be a basis of B and write zk = ni=1 cik bi in the basis B. There exists at least one i0 ∈ {1, 2, . . . , n} such that the sequence ci0 k is unbounded; otherwise, zk would be bounded. One can slightly modify the basis B so that this condition is satisfied for all i. In fact, assume that cik is unbounded for 1 ≤ i ≤ m < n and cik is bounded for m + 1 ≤ i ≤ n. Take then the new basis vectors n

b1 = b1 −

bi , bi = bi , when i ≥ 2.

i=m+1

In the new basis, one can express the vector zk as zk =

m i=1

cik bi +

n

(cik + c1k )bi ,

i=m+1

and all components are now unbounded. k = {x; x = nWe consider now the intervals Jik with the property that U k i=1 αik bi , with αik ∈ Jik } is a neighborhood of zk and f (x) > 2 , for all x ∈ Uk . The argument for n = 1 shows that there exists a sequence of integers (mik ) going to infinity such that we have a nested sequence of nontrivial compact intervals ··· ⊂ Take

1 1 Jik+1 ⊂ Jik ⊂ · · · ⊂ Ji1 . mik+1 mik ∞ 1 1 γi ∈ Jik mik k=1

+ = {γi bi , i ∈ {1, 2, . . . , n}}. It follows that f is unbounded and construct a new basis B on EB+. Problem 4.47. It is known that if f, g : C → C are entire functions without common zeros then there exist entire functions a, b : C → C such that a(z)f (z)+b(z)g(z) = 1 for all z ∈ C. 1. Prove that we can choose a(z) without any zeros. 2. Is it possible to choose both a and b without zeros?

214

8 Analysis Solutions

Solution 4.47. 1. We know that Af + Bg = 1 for some entire functions A and B. Therefore, for any entire function λ, we have also (A + λg)f + (B − λf )g = 1. We now choose a = A+λg such that A+λg = eh . If z0 is a zero of g, we need to have eh(z0 ) = A(z0 ) (with the same multiplicity). This is possible from the interpolation theorem for entire functions, because A(z0 )f (z0 ) = 1. 2. Let f, g ∈ C[z] be nonconstant polynomials of different degrees. We will show that there are no entire functions a, b such that af + bg = 1 and a, b have no zeros. There are no integer nonconstant functions. First, there exist nonzero polynomials A and B such that Af + Bg = 1. As above, for any entire function λ : C → C, one can construct other solutions, namely a = A − λg and b = B + λf . Let us now show that any pair a, b can be obtained in this way. We have (A − a)f + (B − b)g = 0. Thus the set of zeros of (A − a) contains the set of zeros of g (since f and g have no common zeros). In particular, there exists an entire function λ such that A − a = λg. In a similar way, there exists an entire function μ such that B − b = μf . The condition above shows that λ = μ, as claimed. If a, b do not have zeros, then the meromorphic function F = λ1 must be distinct everywhere from the rational functions a1 , a2 , a3 given below: a1 (z) = 0,

a2 (z) = g(z)/A(z),

a3 (z) = −f (z)/B(z).

Consider the meromorphic function G=

F − a1 a3 − a2 · . F − a 2 a3 − a 1

The points where G(z) ∈ {0, 1, ∞} are among those satisfying one of the conditions a1 (z) = a2 (z), a1 (z) = a3 (z), a2 (z) = a3 (z), and therefore they are finitely many. According to Picard’s theorem, G does not have an essential singularity at ∞ and thus G is rational. This implies that F is rational; hence λf is a polynomial and thus a and b are polynomials. But polynomials without zeros are constant, and this is impossible since f and g have different degrees. Problem 4.48. Consider a compact set X ⊂ R. Show that a necessary and sufficient condition for the existence of a monic nonconstant polynomial with real coefficients h ∈ R[x] such that |h(x)| < 1 for all x ∈ X is the existence of monic nonconstant polynomial g(x) ∈ R[x] such that |g(x)| < 2 for all x ∈ X. Prove that 2 is the maximal number with this property. Solution 4.48. Let P(X) be the Banach space of real polynomials endowed with the norm θ = supx∈X |θ (x)|. We define the nonlinear operator A : P(x) → P(x) given by

8 Analysis Solutions

215

1 (Af (x)) = f 2 (x) − f 2 , for f ∈ P(X). 2 If f is monic, then Af is also monic and we have Af = 21 f 2 . An easy induction implies that 2f 2n An f = . n 22 Moreover, if f < 2, then there exists some n for which An f < 1. This proves the first part. Let X = [−2, 2]. The norm of p(x) = x on X is p = 2. We will prove that any monic polynomial on X has norm at least 2. Assume the contrary, namely that there exists f ∈ P(x), monic, with f < 2. We have two intermediate results: 1. There exists an operator T : P(X) → P(X) that takes monic polynomials of degree 2k to monic polynomials of degree k such that T P < P . 2. Assuming the existence of a monic polynomial of norm f < 2, there exists n ∈ Z+ and a monic polynomial g, of degree 2n , with g < 2. By points 1 and 2 above there exists a monic polynomial g of degree 2n with g < 2 on [−2, 2], which implies that there exists a monic polynomial of degree 1 such that g < 2 on [−2, 2], which is obviously false. ˜ Proof of claim 1. If h ∈ P(x), let us consider h(x) = 21 (h(x) + h(−x)), which ˜ is even and thus can be written as a polynomial in x 2 . Thus h(x) = Q x 2 , where now Q : [0, 4] → R. Define T (h)(x) = Q(x + 2). It is clear that T has the desired properties. Proof of claim 2. If f has degree 2m p, where p is odd, then t = T m f is monic of degree p, t < 2. There exists n ∈ Z+ such that p divides 2n −1, and so 2n −1 = pq. We define g ∈ P(x) by means of g = t (x)q x. It is immediate that g < 2, and the proof is complete.

9 Glossary

9.1 Compendium of Triangle Basic Terminology and Formulas 9.1.1 Lengths in a Triangle We used the standard notation for the important features of a triangle, as follows: • • • • • • • • • •

a, b, c the sides lengths A, B, C the angles ma , mb , mc the medians ha , hb , hc the altitudes wa , wb , wc the angle bisectors R the radius of the circumcircle r the radius of the incircle ra , rb , rc the radii of the extrinsic circles S the area p the semiperimeter

We collect below some useful identities between these quantities: • • • • • • • • • • •

S = 21 bc sin A = (p(p − a)(p − b)(p − c))1/2 = 4Rr cos A2 cos B2 cos C2 , S = rp, 4RS = abc, ra (p − a) = rp = S, ra = p tan A2 , 4R + r = ra + rb + rc , aha = 2S, 3 2S 2 ha hb hc = 8S abc = R , 1 1 1 1 1 1 1 ha + hb + hc = ra + rb + rc = r , m2a + m2b + m2c = 43 a 2 + b2 + c2 , √ √ bc p(p − a), wa = 2 b+c " p(p−b)(p−c) , ra = p−a

218

9 Glossary

•

2 2 a 2 + b2 + " c = 8R (1 + cos A cos B cos C),

•

sin

• •

cos A + cos B + cos C = 1 + 4 sin " tan A2 = (p−b)(p−c) p(p−a) ,

• •

a b c R = 2 sin A = 2 sin B = 2 sin C , 2 2 2 2 4ma = 2b + 2c − a .

A 2

=

(p−b)(p−c) , bc

A 2

sin

B 2

sin

C 2

=1+

r R,

9.1.2 Important Points and Lines in a Triangle • • • • •

•

•

•

•

• •

•

•

The circumcenter O is the center of the triangle’s circumcircle. It can be found as the intersection of the perpendicular bisectors of the sides. The incenter I is the center of the incircle and thus the intersection of the angle bisectors. The orthocenter H is the intersection of the three altitudes of the triangle. The centroid G is the intersection of the three medians. outside the The excenter JA is the center of the excircle lying in the angle BAC, triangle ABC and tangent to the lines AB, AC and to the segment BC. There are three excenters, JA , JB , JC , using similar definitions. If MA , MB , MC are the midpoints of BC, CA, AB respectively, then the three lines MA JA , MB JB , MC JC are concurrent, and their intersection point is called the mittenpunkt. If SA , SB , SC are the contact points of the excircles with BC, CA, AB respectively, then ASA , BSB , CSC are concurrent and their intersection is called the Nagel point N a. Let TA , TB , TC be the contact points of the incircle with sides BC, CA, AB respectively. Then the lines ATA , BTB , CTC are concurrent, and their intersection point is called the Gergonne point Ge. The isogonal conjugate of the point X is the intersection of the three concurrent lines that are the reflections of AX, BX, CX about the angle bisectors of A, B, C respectively. The intersection point of the symmedians is called the symmedian point (or the Lemoine point). It is the isogonal conjugate of the centroid G. Two lines passing through the vertex A of a triangle ABC are called isotomic if they cut the side BC in points symmetric with respect to the midpoint of BC. Given a point P , the isotomic lines of the Cevians AP , BP , and CP meet at a point P called the isotomic point of P . The Fermat point F of an acute triangle is the point that minimizes the sum of the distances to the vertices. If we draw equilateral triangles BCA , ABC , ACB on the outside of the triangle ABC, then the three lines AA , BB , CC are concurrent at F . The Euler line: the points O, G, H are collinear and |H G| : |GO| = 2 : 1.

9.1 Compendium of Triangle Basic Terminology and Formulas

• •

219

The Gergonne point Ge, the triangle centroid G, and the mittenpunkt M are collinear, with |Ge G| : |GM| = 2 : 1. The Nagel line: the points I, G, N a are collinear and |N a G| : |GI | = 2 : 1.

Kimberling tabulated and enumerated properties of more than 1477 centers in a triangle (later Brisse extended the list to 2001 items). The interested reader might consult his updated web page: •

C. Kimberling: see http://faculty.evansville.edu/ck6/encyclopedia/ETC.html

9.1.3 Coordinates in a Triangle We are given a reference triangle ABC. Barycentric coordinates. The barycentric coordinates of the point P are given by an ordered triple FP = (tA , tB , tC ) (up to nonzero scalar multiplication) such that P is the centroid of the system consisting of the three masses tA , tB , tC located at the respective vertices A, B, C of the triangle, i.e., the mass tA is located at A, etc. These coordinates were introduced by Möbius in 1827. The coordinates tA are proportional to the areas area(P BC). If they are actually equal to these areas, they are called homogeneous barycentric coordinates. Here are the coordinates of the principal points in a triangle. • • • • • • •

FG = (1, 1, 1) FO = a 2 b2 + c2 − a 2 , b2 a 2 + c2 − b2 , c2 a 2 + b2 − c2 FJA = (−a, b, c), FI = (a, b, c) FGe = ((p − b)(p − c), (p − c)(p − a), (p − a)(p − b)) FNa = (p− a, p − b, p −c) a 2 + c2 − b2 a 2 + b2 − c2 , a 2 + b2 − c2 b2 + c2 − a 2 , FH = 2 a + c2 − b2 b2 + c2 − a 2 FK = a 2 , b 2 , c 2

The line determined by the points (s1 , s2 , s2 ) and (t1 , t2 , t3 ) has the equation ⎛ ⎞ s 1 s2 s3 det ⎝ t1 t2 t3 ⎠ = 0. x1 x2 x3 Trilinear coordinates. The trilinear coordinates of a point P are given by an ordered triple of numbers proportional to the directed distance from P to the sides, up to multiplication by a nonzero scalar. If we normalize them so that they give the distances to the sides, then these are called the exact trilinear coordinates. Trilinear coordinates are also known as homogeneous coordinates. They were introduced by Plücker in 1835. For instance, the vertices have trilinear coordinates (1 : 0 : 0), (0 : 1 : 0), and (0 : 0 : 1).

220

9 Glossary

For most important points in a triangle, the trilinear coordinates are given by triples of the form (f (a, b, c) : f (b, c, a) : f (c, a, b)), for some function f on the sides a, b, c. For instance, we have fH (a, b, c) = cos B cos C, fG (a, b, c) = a1 , fO (a, b, c) = cos A, fI (a, b, c) = 1. For the Fermat point, we can take fF (a, b, c) = sin(A + π3 ). The homogeneous barycentric coordinates corresponding to (x : y : z) are (ax, 1 1by,1 cz). The trilinear coordinates of the isogonal conjugate of (x : y : z) are x, y, z .

9.2 Appendix: Pell’s Equation Pell’s equation is the Diophantine equation x 2 − dy 2 = 1. The first treatment of this equation was actually given by Lord William Brouncker in 1657, but his solution was erroneously ascribed to John Pell by Euler. Lagrange developed the theory of continued fractions, which gives the actual method of finding the minimal solution, and published the first proof in 1766. However, some methods (the cyclic method) were already known to the Indian mathematicians Brahmagupta and Bhaskara in the twelfth century. General solutions. Pell’s equation has infinitely many solutions if d is not a perfect square and none otherwise. If (x0 , y0 ) denotes its minimal solution (called also the fundamental solution) different from (1, 0), then all solutions (xn , yn ) are obtained from it by means of the recurrence relations xn+1 = x0 xn + dy0 yn ,

yn+1 = y0 xn + x0 yn ,

x1 = x0 , y1 = y0 .

The fundamental solution. Thus the main point is finding the fundamental solution. This can be achieved using continued fractions. According to Lagrange, the continued fraction of a quadratic irrational number (i.e., a real number satisfying a quadratic equation with rational coefficients that is not rational itself) is periodic after some point; in our case, even more can be obtained, namely

√ d = a0 , a1 , . . . , am , 2a0 . The numbers aj can be calculated recursively, as follows. Set a0 = the sequences Q0 = 1, P0 = 0, P 1 = a0 , Q1 = d − a02 , Pn = an−1 Qn−1 − Pn−1 , Qn = Then

a0 + Pn an = . Qn

d−Pn2 Qn−1 .

√ d and construct

9.2 Appendix: Pell’s Equation

221

Furthermore, consider the sequences p0 = a0 , q0 = 1, p1 = a0 a1 + 1, q1 = a1 , pn = an pn−1 + pn−2 , qn = an qn−1 + qn−2 . Then we have the identities pn2 − dqn2 = (−1)n+1 Qn+1 , which permit one to construct solutions to the Pell-type equations. If am+1 = 2a0 , as happens for a quadratic irrational, then Qm+1 = 1. In particular, we find that the fundamental solution is determined by the parity of m, as follows: x0 = pm , y0 = qm , if m is odd, x0 = p2m+1 , y0 = q2m+1 , if m is even. Observe that pr /qr = [a0 , a1 , . . . , ar ]. The general solution (xn , yn ) is obtained by means of the trick x 2 − dy2 = (x02 − dy02 )n = 1, and setting x+

√ √ n dy = x0 + dy0 ,

we obtain the family of solutions √ √ n n x0 + dy0 + x0 − dy0 , xn = 2

x−

√

√ n dy = x0 − dy0 ,

yn =

x0 +

√

dy0

n

√ n − x0 − dy0 . 2

The negative Pell equation. The Pell-like equation x 2 − dy 2 = −1 can also be solved, but it does not always have solutions. When it has one, then it has infinitely many. A necessary condition for this equation to be solvable is that all odd prime factors of d be of the form 4k + 1, and that d ≡ 0 (mod 4). However, these conditions are not sufficient, as can be seen from the equation for d = 34, which has no solutions. The method of continued fractions works for these equations as well. In fact, for this Pell-type equation, we have the fundamental solution x0 = pm ,

y0 = qm ,

if m is even.

The equation does not have solutions if m is odd. Furthermore, the general solution is given again by √ √ √ √ n n n n x0 + dy0 + x0 − dy0 x0 + dy0 − x0 − dy0 , yn = , xn = 2 2

222

9 Glossary

but only for odd n. Pell-like equations. The Pell-like equation x 2 − dy 2 = c √ can be solved using the same ideas. If |c| < d, then the equation has solutions if and only if ' 8 c ∈ Q0 , −Q1 , . . . , (−1)m+1 Qm+1 . √ Moreover, if c > d, then the procedure is significantly more complicated. See for instance the comments of Dickson on this subject. It is clear that using the solutions (xn , yn ) to the companion Pell equation x 2 − dy 2 = 1 and a particular solution (u, v) of the general Pell-type equation from above, we are able to find infinitely many solutions by means of the recurrence un = xn u ± dyn v,

vn = xn v ± yn u.

However, even if we start with the minimal solution, we cannot always obtain all solutions of the equation using this recurrence. The reason is that we might well have several fundamental solutions. For instance, if d = 10 and c = 9, then we have the fundamental solutions (7, 2), (13, 4) and (57, 18). The general equation ax 2 − by 2 = c can be reduced to the former equation by setting d = ab, x = ax, c = ac and looking for those solutions where x is divisible by a. The degree-two equation in two unknowns. Moreover, the general degree-two equation ax 2 + bxy + cy 2 + dx + cy + f = 0 where a, b, c, d, e, f ∈ Z, can be reduced to either Pell-type equations or elliptic equations. 1. The case = b2 − 4ac < 0. The equation represents an ellipse in the plane, and therefore it has only finitely many integer solutions. These equations were solved by Gauss and earlier considered by Euler. According to Dickson one can find all its solutions using the solutions to Pell’s equations. 2. If = 0, the parabolic-type equation becomes 1 (2ax + by + d)2 + 2(ae − bd)y + 4af − d 2 = 0. 4a If 2ae − bd = 0, then the equation is equivalent, through a linear transformation, to (2ax + by + d)2 = d 2 − 4af ; hence the parabola degenerates into two lines. If d 2 − 4af isa perfect square, then we have two infinite families of solutions 2ax + by + d = ± d 2 − 4af ; otherwise, there are no integer solutions.

9.2 Appendix: Pell’s Equation

223

If 2ae−bd = 0, then we have to solve the equation 2(ae−bd)y = −t 2 +d 2 −4af , for an arbitrary parameter t ∈ Z. This depends only on computing the residues modulo 2(ae − bd) of the perfect squares, and yields infinitely many solutions if one exists. Furthermore, 2ax = t − d − by determines x. 3. If > 0, then the hyperbolic-type equation becomes: a(x − 2cd + bc)2 + b(x − 2cd + bc)(y + 2ac + bd) + c(y − 2ac + bd)2 = − ac2 + cd 2 + ef 2 − bdc − 4acf , which, through a transformation u = x − 2cd + bc, v = y − 2ac + bd, becomes au2 + buv + cv 2 = h. This equation can be reduced to a Pell equation. It will have then either zero or infinitely many solutions. The Ankeny–Artin–Chowla Conjecture. Although the Pell equation seems to be well understood, there are many subtle questions concerning its solutions. Here is a problem that has resisted to all efforts until now. 1. If p is a prime p ≡ 1 (mod 4) and y0 is the smallest positive value of y such that x 2 − py 2 = −4, then y0 ≡ 0 (mod p). 2. If p is a prime p ≡ 3 (mod 4) and y0 is the smallest positive value of y such that x 2 − py 2 = 1, then y0 ≡ 0 (mod p). The first conjecture is due to Ankeny, Artin, and Chowla, and the second to Mordell. The Thue theorem for higher degree. The situation of equations of higher degree is completely different. Let f (x, y) = a0 x n + a1 x n−1 y + a2 x n−2 y 2 + · · · + an y n be a homogeneous form of degree n. If n ≥ 3 and f is irreducible over Q, then the equation f (x, y) = g(x, y), where g(x, y) is an arbitrary form of degree n − 2 (not necessarily homogeneous), has only finitely many solutions. This is a celebrated theorem due to Thue and based on results of Siegel and Roth. Later A. Baker gave explicit bounds on the size of the solutions for a wide class of such f . •

•

L.E. Dickson: Pell equation: ax 2 + by 2 + c made square, Chapter 12 in History of the Theory of Numbers, Vol. 2: Diophantine Analysis, New York, Chelsea, 341–400, 1952. H.W. Lenstra, Jr.: Solving the Pell equation, Notices Amer. Math. Soc. 49 (2002), 2, 182–192.

Index of Mathematical Results

ˇ Cebotarev density theorem, 83 Ankeny–Artin–Chowla conjecture, 223 Birch–Swinnerton-Dyer conjecture, 81 Blaschke theorem, 211 Bonnesen’s isoperimetric inequality, 170 Bonsé–Pósa inequality, 46 Borsuk problem, 153, 156, 175 Borsuk–Lusternik–Shnirelman theorem, 175 Catalan’s conjecture, 56 Cauchy’s integral formula, 167 Cauchy’s rigidity theorem, 171, 173 Cauchy–Schwartz inequality, 167 Cauchy–Schwarz inequality, 175 Ceva’s theorem, 139 Chebyshev theorem, 40, 45, 46, 58 Chinese Remainder theorem, 82 Chinese remainder theorem, 50, 78 congruent numbers problem, 80 De Moivre’s identity, 60 Dirac theorem, 97 Dirichlet theorem on arithmetic progressions, 50, 82, 97 Eisenstein criterion, 38 Erd˝os conjecture, 114 Erd˝os theorem, 126 Erd˝os–Straus conjecture, 63 Euler conjecture, 42 Euler criterion, 72, 73 Euler theorem in a triangle, 182

Fermat equation, 42, 67 four color theorem, 104 Frobenius coin problem, 50 fundamental theorem of symmetric polynomials, 85, 89 Gowers theorem, 123 Green–Tao theorem, 123 Hadamard inequality, 186 Hahn–Banach lemma, 131 Heawood conjecture, 103 Helly’s theorem, 159 intermediate value theorem, 188 isoperimetric inequality, 147 Jensen inequality, 179 Jung’s theorem, 159 Kahn–Kallai theorem, 156 Kaprekar constant, 95 Lagrange four squares theorem, 74 Lagrange mean value theorem, 194 Lagrange multiplier method, 189 Lagrange theorem on periodic continued fractions, 220 Laplace method, 198 Lehmer conjecture, 57 Ljunggren equation, 69 Maclaurin series, 189

226

Index of Mathematical Results

maximum principle for holomorphic functions, 211 Means inequality, 161, 179, 180, 189 Millennium problems, 81, 195 Minkowski’s theorem, 165

Roché inequality, 28, 182 Sturm comparison theorem, 212 superperfect numbers conjecture, 60 Sylvester’s problem, 172 Szemerédi theorem, 122

odd perfect number conjecture, 59 Pell equation, 43 Pell’s equation, 220 Picard theorem, 214 pigeonhole principle, 86, 92, 96, 97, 108, 112

Taylor formula, 209 Thales’ theorem, 140 Thue’s theorem, 223 Ulam problem, 126 Van der Waerden theorem, 122

quadratic reciprocity theorem, 50 Rayleigh–Beatty theorem, 101 Riemann hypothesis, 195

Waring problem, 52, 74, 76 Wiles theorem, 42 Wilson theorem, 44, 53

Index of Mathematical Terms

abundant number, 9, 81 altitude, 22, 25, 27, 139, 160, 162, 175, 217 annulus, 190 area, 8, 19, 23, 25–27, 70, 80, 124, 140–142, 145, 154, 157, 159, 161, 165–167, 170, 173, 176, 177, 181, 182, 190, 217 arithmetic progression, 4, 7, 12, 40, 50, 56, 82, 87, 97, 122 asymptotic, 16, 78, 106, 114, 128 ball, 25, 27, 143, 156–159, 174, 175 Banach space, 214 barycentric coordinates, 142, 219 Bernoulli numbers, 54 Bessel function, 210 binomial binomial coefficient, 14, 41, 62, 97, 98, 105, 109 binomial expansion, 38, 62, 104 binomial coefficient, 3, 37 bisector, 21, 25, 27, 133, 134, 162, 175, 217 body, 25, 145, 156–158, 166 boundary, 18, 25, 150, 151, 156–158, 161, 165, 166, 174, 211 cardinal of a set, 15, 17, 50, 78, 88, 92, 100–102, 104, 108, 111, 112, 117, 128, 191 cardinality of a set, 15 centroid, 218 Cevian, 139 choice number, 3, 37 chord, 21, 22, 24, 135–137, 152, 157, 159, 160

circular permutation, 62, 190 circumcenter, 218 circumscribable, 23, 140 collinear points, 23, 126, 138, 140–142, 172, 182, 218, 219 commensurable, 126, 203 compact, 34, 151, 157, 212–214 complex number, 13, 24, 30, 89, 92, 126, 135, 136, 147, 148, 190, 200 complex plane, 55 computer search, 43, 63, 65 computer-assisted proof, 104 concave, 149 congruence modulo p, 3, 9, 39, 40, 44, 50, 52, 61, 63, 66, 72, 73, 78–80, 82, 83 congruence transformation, 24, 151 congruent number, 80 continued fraction, 220 convergent, 9, 30–33, 77, 78, 190, 191, 193, 194, 203, 205 convex, 16, 22–26, 106, 136, 139, 143, 144, 147, 149, 156–158, 167, 168, 170, 171 convex hull, 136, 156, 158 covering, 27, 173 critical, extremal point, 180, 190 deficient number, 9, 81 dense, 34, 126, 131, 142, 157, 203, 211 density, 83, 109, 122, 128 determinant, 11, 86, 87, 109, 138, 141, 186 diameter, 24, 25, 145, 147, 151, 152, 154, 156, 158, 167, 174 differential equation, 198 digits, 4, 5, 13, 40, 47, 95

228

Index of Mathematical Terms

Diophantine equation, 42, 53, 56, 64, 67, 69, 70, 80, 115, 220 discriminant, 48, 83, 194 distance, 16, 19, 20, 22–24, 26, 107, 122, 126, 128, 130, 131, 136, 142, 143, 145, 146, 148, 151, 152, 165, 167, 169, 219 domain, 26, 170, 171, 174 eigenvalue, 90, 91, 211 Euler inequality, 28, 177 Euler line, 22, 138, 218 Euler totient function, 3, 6, 7, 39, 51, 57, 61, 92 Euler–Poincaré characteristic, 103, 150, 171, 172 Eulerian circuit, 19, 123 excenter, 218 factorial, 3, 37 Fermat point, 22, 138, 218 Fibonacci sequence, 118, 185 fractional part, 193 function, 12, 23, 31, 33, 88, 144, 165, 194, 204, 207 additive function, 208 bijective function, 15, 102 concave (or convex) function, 179 continuous function, 33, 34, 71, 153, 157, 206, 209, 210 decreasing function, 31, 139, 196 entire function, 34, 213 function, 104 generating function, 16, 106 holomorphic function, 211 hyperbolic trigonometric function, 203 increasing function, 15, 64, 101, 110, 189 meromorphic function, 214 periodic function, 31, 193 positive function, 32, 198 rational function, 13, 42, 92 smooth function, 32–34, 87, 162, 198, 199, 208, 212 Gauss map, 156 Gergonne point, 23, 141, 218, 219 graph, 15, 19, 97, 102, 103, 123, 147, 150 Hadwiger number, 163 Hamiltonian cycle, 97

homographic division, 138 homothety, 26, 106, 126, 136, 170, 171 incenter, 218 index of mathematical terms, 37 infinite descent method, 56, 67 inscribable, 8, 22, 23, 26, 70, 137, 140, 141, 169 integer part, 4, 39 integral equation, 34, 209 integral operator, 211 intrinsic perimeter, 164 isosceles, 18, 21, 25, 122, 134, 135, 138, 159, 160 isotomic point, 218 Jacobian matrix, 190 lattice, 26, 122, 129, 165 Legendre symbol, 72, 82 Lemoine (symmedian) point, 218 length, 19, 21–27, 29, 94, 96, 124, 125, 135, 137, 140, 143, 145, 146, 151, 152, 154–156, 159, 160, 162, 166–169, 171, 175, 182, 187, 211, 217 linear linear combination, 88, 93, 209 linear equation, 113 linear function, 33, 86, 204, 207 linear functional, 131 linear independence, 13, 25, 91, 92, 162 linear map, 131 matrix, 26, 86, 87, 91, 109, 111, 141, 142, 162, 164, 186, 190 median, 27, 175, 217 Mersenne prime number, 55 Mersenne primes, 59 Minkowski norm, 164 mittenpunkt, 218 Nagel line, 219 Nagel point, 23, 141, 218 nilpotent, 90, 93 non-collinear points, 19 noncollinear points, 27, 124, 126, 172 nonlinear operator, 214 number of combinations, 3, 37 orthocenter, 218

Index of Mathematical Terms orthogonal, 130, 136, 137, 143, 145, 153–157 packing, 19, 125, 129 parameterization (of a curve), 41 partition, 15, 17, 26, 99, 103, 113, 114, 151, 153, 156, 158, 168, 169, 171, 174 pentagonal number, 8, 69 perfect number, 59 periodic, 31, 193, 203, 225 poder, 181 polygon, 16, 23, 26, 29, 106, 121, 143, 144, 146, 147, 150, 158, 168, 171, 187 polyhedron, 27, 153, 157, 171–173 polynomial characteristic polynomial, 91 irreducible polynomial, 3, 38, 83, 92, 223 monic polynomial, 34, 214 polynomial with arbitrary coefficients, 11, 13, 86, 91, 93, 211, 214 polynomial with integer coefficients, 6, 9, 12, 40, 51, 61, 82, 89 real polynomial, 12, 29, 30, 33, 86, 87, 185, 188, 206, 208 set of polynomials, 32, 197 symmetric polynomial, 85, 89 projective duality, 172 projective transformation, 138 Pythagorean triple, 5, 42, 44, 66, 67, 69 quadratic irrational number, 220 quadratic residue, 50, 68, 72, 73

229

, 29, 185 relatively prime, 39, 40, 46, 65, 78, 86, 93 Riemann zeta function, 54, 194 root, 34, 61, 83, 89, 91–93, 160, 200, 212 sequence, 4, 6, 9, 12, 14, 17, 18, 30–32, 40, 43, 48, 50, 61, 74, 76, 77, 81, 87, 95, 96, 98, 100, 110, 112, 114, 116, 117, 119, 127, 128, 147, 191, 193, 196, 201, 203, 206, 207, 213 series, 77 simplex, 25, 156, 158, 159, 171 stereographic projection, 41 sum-free set, 17, 114 superperfect number, 60 surface, 26, 103, 154, 166, 174 symmedian, 181 symmetric, 137, 154, 155, 161, 166, 174, 175 symmetry, 51, 88, 97, 106, 133, 136, 137, 160, 166, 170, 172, 205 tiling, 121, 149 torus, 13, 93 transcendental number, 55, 128 triangular number, 8, 69 trilinear coordinates, 138, 141, 219 vector space, 90, 130, 208 width, 23, 145, 152, 153, 155, 166, 167

Index of Topics and Methods

Classical arithmetic functions and ap- Arithmetic progressions: 1.10, 1.26, plications: 1.4, 1.29, 1.34, 1.36, 1.42, 1.40, 2.5, 2.21 1.43, 1.44, 1.45, 1.46, 1.69 Representing integers by algebraic Congruences and divisibility: 1.1, 1.3, functions, Waring problem: 1.31, 1.53, 1.8, 1.11, 1.15, 1.17, 1.22, 1.24, 1.27, 1.61, 1.62, 1.63, 1.64 1.38, 1.48, 1.50, 1.51, 1.67, 1.68, 2.7, 2.45 Rational and algebraic numbers: 1.47, Algebraic number theory: 1.70 2.14, 2.15 Groups: 2.10 Polynomials: 1.2, 1.7, 1.30, 2.2, 2.4, 2.9, 2.12, 2.13, 2.24 Combinatorial identities: 2.33, 2.34 Estimates of arithmetically defined Algebraic equations: 2.3, 2.16 functions: 1.5, 1.6, 1.23, 1.28, 1.37, 1.55, 1.65, 1.66, 2.25, 2.35, 2.47 Linear algebra, matrices, determinants: 2.6, 2.11, 2.17, 2.18, 2.40, 2.43 Symmetric functions: 2.1, 2.8 General Diophantine equations (Pyth- Combinatorial properties of sets of inagorean, exponential): 1.12, 1.14, 1.16, tegers: 2.26, 2.27, 2.28, 2.29, 2.41, 2.44 1.20, 1.25, 1.33, 1.35, 1.39, 1.41, 1.49, 1.52, 1.56, 1.57, 1.60 Sequences of integers: 2.19, 2.42, 2.52, 2.53 Pell’s equation and its applications: 1.13, 1.58, 1.59 Counting problems: 2.23, 3.21 Continued fractions: 1.13 Extremal problems concerning sets of Prime numbers, Chebyshev theorem: integers: 2.36, 2.37, 2.48, 2.49, 2.50, 1.9, 1.18, 1.19, 1.21, 1.32, 1.54, 2.46 2.51

232

Index of Topics and Methods

Ramsey-type problems: 2.20, 2.30, 2.32, 2.38, 2.39, 2.56, 2.57, 2.59, 2.62

Geometric inequalities in the triangle: 3.62, 3.63, 3.64, 3.65, 3.66, 3.67, 3.68, 3.69, 3.70, 3.71, 3.72, 3.73

Graphs and applications: 2.22, 2.31, 2.58 Topological methods in geometry: 3.32, 3.33, 3.34, 3.35, 3.36, 3.38, 3.39, 3.58, Tiling, packing and covering: 2.55, 3.59, 3.61 2.60, 2.64, 3.46, 3.47, 3.53

Area and volume: 3.42, 3.45, 3.51, 3.56, General inequalities: 4.2, 4.3, 4.4, 4.6, 4.14, 4.15, 4.22, 4.26 3.60 Distances in combinatorial geometry: Convexity of real functions: 4.16 2.61, 2.63, 3.23, 3.48, 3.54 Geometry of the triangle: 3.1, 3.11, 3.12, 3.14, 3.16 Complex numbers and applications: 3.5, 3.6, 3.8, 4.1, 4.19, 4.34

Minima and maxima of real functions, mean values: 4.7, 4.8, 4.9, 4.10, 4.37 Functions of a complex variable, holomorphic functions: 4.20, 4.24, 4.44, 4.47

Geometric constructions, geometric loci: 3.2, 3.3, 3.4, 3.7, 3.9, 3.10, 3.15 Sums of series and convergence of sequences: 4.11, 4.12, 4.17, 4.18, 4.23, Trigonometry: 3.17, 3.30 4.28, 4.29, 4.31 Analytic methods in geometry: 3.18 Dynamical systems, density of sets of Higher dimensional geometry: 3.13, reals: 3.19, 4.32, 4.35, 4.43 3.57 Integrals: 4.13, 4.21, 4.30 Convex geometry, integral formulas, isoperimetric inequalities: 3.24, 3.25, Functional equations and inequations: 3.29, 3.49, 3.50, 3.55 4.27, 4.33, 4.36, 4.38 Extremal problems concerning finite sets of points: 2.54, 3.20, 3.27, 3.28, Integral and differential equations: 4.25, 4.39, 4.40, 4.41, 4.42, 4.45 3.31, 3.37, 3.40, 3.41, 3.44, 3.52 General geometric inequalities: 3.22, Linear functionals and operators: 2.65, 3.26, 3.43, 4.5 4.46, 4.48

The Math Problems Notebook

Birkh¨auser Boston • Basel • Berlin

Valentin Boju MontrealTech Institut de Technologie de Montreal P.O. Box 78575, Station Wilderton Montreal, Quebec, H3S 2W9 Canada [email protected]

Louis Funar Institut Fourier BP 74 URF Mathematiques Universit´e de Grenoble I 38402 Saint Martin d’Heres cedex France [email protected]

Cover design by Alex Gerasev. Mathematics Subject Classification (2000): 00A07 (Primary); 05-01, 11-01, 26-01, 51-01, 52-01 Library of Congress Control Number: 2007929628 ISBN-13: 978-0-8176-4546-5

e-ISBN-13: 978-0-8176-4547-2

Printed on acid-free paper. c 2007 Birkh¨auser Boston All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Birkh¨auser Boston, c/o Springer Science+Business Media LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights.

9 8 7 6 5 4 3 2 1 www.birkhauser.com

(TXQ/SB)

The Authors

Valentin Boju was professor of mathematics at the University of Craiova, Romania, until his retirement in 2000. His research work was primarily in the field of geometry. He further promoted biostatistics and biomathematics as a discipline within the Department of Medicine of the University of Craiova. He left Romania for Canada, where he has taught at MontrealTech since 2001. He was actively involved in coaching college students in problem-solving and intuitive mathematics. In 2004, he was honored with the title of Officer of the Order “Cultural Merit,” Category “Scientific Research.” MontrealTech Institut de Technologie de Montréal P.O. Box 78575 Station Wilderton Montréal, Quebec, H3S 2W9 Canada e-mail: [email protected] Louis Funar has been a researcher at CNRS, at the Fourier Institute, University of Grenoble, France, since 1994. During his college years, he participated three times in the International Mathematics Olympiads with the Romanian team in the years 1983–1985, winning bronze, gold, and silver medals. His main research interests are in geometric topology and mathematical physics. Institut Fourier, BP 74 Mathématiques UMR 5582 Université de Grenoble I 38402 Saint-Martin-d’Hères cedex France e-mail: [email protected]

To all schoolteachers, particularly to my wonderful parents, Evdochia and Nicolae Boju, who genuinely believed in the education of the younger generation.

To my parents, Maria and Ioan Funar, who laid the foundation for our collaboration, as a late reward for their hard work and help, for their enthusiasm, determinedness, and strength during all these years.

Preface

The authors met on a Sunday morning about 25 years ago in Room 113. One of us was a college student and the other was leading the Sunday Math Circle. This circle, within the math department of the University of Craiova, gathered college students who possessed a common passion for mathematics. Most of them were participating in the mathematical competitions in vogue at that time, namely the Olympiads. Others just wanted to have good time. There were similar math circles everywhere in the country, in which high-school students were committed to active training for math competitions. The highest standard was achieved by the selective training camps organized at the national level, which were led by professional coaches and mathematicians who were able to stimulate and elicit high performance. To mention a few among the leaders, we recall Dorin Andrica, Toma Albu, Titu Andreescu, Mircea Becheanu, Ioan Cuculescu, Dorel Mihet, and Ioan Tomescu. The Sunday Math Circle shifted partially from its purpose of Olympiad training by following freely the leader’s ideas and thus becoming a place primarily intended for disseminating beautiful mathematics at an elementary level. The fundamental texts were the celebrated book of Richard Courant and Herbert Robbins with the mysterious title What is mathematics?, and the book of David Hilbert and Stefan Cohn–Vossen entitled Geometry and the Imagination. The participants soon realized the differences in both scale and novelty between the competition-type math problems that they encountered every day and the mathematics built up over hundreds of years and engaged in by professional mathematicians. Competition problems have to be solved in a short amount of time, and people compete against each other. These might be highly nontrivial, unconventional problems requiring deep insights and a lot of imagination, but still they have the advantage that one knows in advance that they have a solution. In the real mathematical world, problems often had to wait not merely years but sometimes centuries to be solved. A working mathematician could go for months or years trying to solve a particular problem that had not been solved before and take the risk of bitterly failing. Moreover, the sustained effort needed for accomplishing such a task would have to take into account a delicate balance between the accumulation of knowledge, methods, and tools and

viii

Preface

the creation from scratch of a path leading to a solution. This intellectual adventure is filled with suspense and frustration, since one does not know for sure what one is trying to prove or whether it is indeed true. Real mathematics seemed to many of us to be too far away from competition problems. The philosophy of the Sunday Math Circle was that, in contrast to what we might think, the border between the two kinds of mathematics could be so vague and flexible that at times one could cross it at an early age. The history of mathematics abounds in examples in which a fresh mind was able to find an unexpected solution that specialists had been unable to find. One needs, however, the right training and the unconventional sense of finding the inspiring problem. Eventually, once a solution has been found, mathematicians are then willing to try to understand it even better, and other solutions follow in time, each one simpler and clearer than the previous one. In some sense, once solved, even the hardest problems start losing slowly their aura of difficulty and eventually become just problems. Problems that are today part of the curriculum of the average high-school student were difficult research problems three hundred years ago, solved only by brilliant mathematicians. This phenomenon demonstrates the evolution that language and science have undergone since then. The authors conceived the present book with the nostalgia of the “good old times" of the Sunday Math Circles. We wanted something that carries the mark of that philosophy, namely, a number of challenging math problems for Olympiads with a glimpse of related problems of interest for the mathematician. The present text is a collection of problems that we think will be useful in training students for mathematical competitions. On the other hand, we hope that it might fulfill our second goal, namely, that of awakening interest in advanced mathematics. Thus its audience might range from college students and teachers to advanced math students and mathematicians. The problems in each section are in increasing order of difficulty, so that the reader give some of the problems a try. We wanted to have 25% easy problems concerning basic tools and methods and consisting mainly of instructional exercises. The beginner might jump directly to the solutions, where a concentrate of the general theory and basic tricks can be found. The largest chunk contains about 50% problems of medium difficulty, which could be useful in training for mathematical competitions from local to international levels. The remaining 25% might be considered difficult problems even for the experienced problem-solver. These problems are often accompanied by comments that put the results in a broader perspective and might incite the reader to pursue the research further. The problems serve as an excuse for introducing all sorts of generalizations and closely related open problems, which are spread among the solutions. Some of these are truly outstanding problems that resisted the efforts of mathematicians over the centuries, such as the congruent numbers conjecture and the Riemann hypothesis, which are among seven Millennium Prize Problems that the Clay Mathematics Institute recorded as some of the most difficult issues with which mathematicians were struggling at the turn of the second millennium and offered a reward of one million dollars for a solution to each one. In mathematics the frontier of our knowledge is still open, and it abounds in important unsolved problems, many of which can be

Preface

ix

understood at the undergraduate level. The reader might be soon driven to the edge of that part of mathematics where he or she could undertake original research. We drew inspiration from the spirit of the famous books by R. K. Guy and his collaborators. Nevertheless, our aim was not to build up a collection of open questions in elementary mathematics but rather to offer a journey among the basic methods in problem-solving. Developing intuition and strengthening the most popular techniques in mathematical competitions are equally part of the goal. In the meantime, we offer the enthusiastic reader a brief glimpse of mathematical research, a place where problems yet unsolved long for deliverance. The present collection of problems evolved from a notebook in which the secondnamed author collected the most interesting and unconventional problems that he encountered during his training for mathematical competitions in the 1980s. In the tradition of the Romanian school of mathematical training, he encountered problems inspired by both Russian Olympiads and American Competitions. Some (if not most) of the problems have already appeared elsewhere, especially in Kvant, Matematika v Shkole, American Mathematical Monthly, Elemente der Mathematik, Matematikai Lapok, Gazeta Matematica, and so, in a sense, the collection gained a certain cosmopolitan flavor. We have given the source in the problem section, when known, and more detailed information in the comments within the solutions part. The original set of problems is complemented by more basic exercises, which aim at introducing many of the tricks and methods of which the competitor should be aware. Years later, we followed the destiny of some of the most intriguing questions from the notebook. Some of them led to developments that are well beyond the scope of this book, and we decided to outline a few of them. We have supplied a short glossary containing some less usual definitions and identities in the geometry of triangles and the solution of Pell’s equation. In order to help the reader find his or her way through the book, we have provided both an index concerning all mathematical results needed in the proofs, which are usually stated at the place where they are used first, and an index of mathematical terms at the end of the book. The authors have benefited from discussions, corrections, help, and feedback from several people, whom we want to thank warmly: Dorin Andrica, Barbu Berceanu, Roland Bacher, Maxime Wolff, Mugurel Barcˇau, Ioan Filip, and Simon György Szatmari. We also thank Ann Kostant, Editorial Director, Springer, and Avanti Paranjpye, Associate Editor, Birkhäuser Boston, for their productive suggestions. One of them led to the “Index of Topics and Methods,” which might be useful for a better reception of the book by readers. We are very grateful to Elizabeth Loew, our Production Editor, for her patience and dedication to accuracy and excellence. Finally, we are thankful to David Kramer for his thorough copyediting corrections. Valentin Boju and Louis Funar Montreal, Canada and Saint-Martin-d’Hères, France July 2006

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Part I Problems 1

Number Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

2

Algebra and Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Algebraic Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Geometric Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11 11 13 18

3

Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Synthetic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Combinatorial Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Geometric Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21 21 23 27

4

Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Part II Solutions and Comments to the Problems 5

Number Theory Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

6

Algebra and Combinatorics Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Algebraic Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Geometric Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

Geometry Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 7.1 Synthetic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 7.2 Combinatorial Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

85 85 95 120

xii

Contents

7.3

Geometric Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8

Analysis Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

9

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 Compendium of Triangle Basic Terminology and Formulas . . . . . . . 9.1.1 Lengths in a Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1.2 Important Points and Lines in a Triangle . . . . . . . . . . . . . . . . . 9.1.3 Coordinates in a Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 Appendix: Pell’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

217 217 217 218 219 220

Index of Mathematical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 Index of Mathematical Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Index of Topics and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

Part I

Problems

1 Number Theory

Problem 1.1. Show that we have Cnk ≡ 0 (mod 2) for all k satisfying 1 ≤ k ≤ n − 1 if and only if n = 2β , where β ∈ Z∗+ . Here Cnk denotes the number of combinations, i.e., the number of ways of picking up a subset of k elements from a set of n elements. Known also as the binomial coefficient or choice number and sometimes denoted as n it is given by the formula k n n! Cnk = = k k!(n − k)! where the factorial n! represents n! = 1 × 2 × 3 × · · · × n. Problem 1.2. Let P = an x n + · · · + a0 be a polynomial with integer coefficients. Suppose that there exists a number p such that: 1. p does not divide an ; 2. p divides ai , for all i ≤ n − 1; 3. p 2 does not divide a0 . Then P is an irreducible polynomial in Z[x]. Problem 1.3. Given mi , bi ∈ Z+ , i ∈ {1, 2, . . . , n}, such that gcd(mi , mj ) = 1 for all i = j , there exist integers x satisfying x ≡ bi (mod mi ) for all i. This result is usually known as the Chinese remainder theorem. Problem 1.4. If gcd(a, m) = 1, then it is a classical result of Euler that we have the following congruence: a ϕ(m)+1 ≡ 0 (mod m), where ϕ(m) is the Euler totient function, which counts how many positive integers smaller than m are relatively prime to m. Prove that this equality holds precisely for those numbers a, m such that for any prime number p that divides a, if pk divides m then pk also divides a.

4

1 Number Theory

Problem 1.5. (Kvant) Let p > 2 be a prime number and ak ∈ {0, 1, . . . , p2 − 1} denote the value of k p modulo p 2 . Prove that p−1

ak =

k=1

p3 − p2 . 2

Problem 1.6. If k ∈ Z+ , then show that k 2 + 1 + · · · + k 2 + 2k = 2k 2 + 2k, where [x] denotes the integer part, i.e., the largest integer smaller than x. Problem 1.7. (Amer. Math. Monthly) If n is not a multiple of 5, then P = x 4n + x 3n + x 2n + x n + 1 is divisible by Q = x 4 + x 3 + x 2 + x + 1. Problem 1.8. (Amer. Math. Monthly) Let p be an odd prime and k ∈ Z+ . Show that there exists a perfect square the last k digits of whose expansion in base p are 1. Problem 1.9. Any natural number greater than 6 can be written as a sum of two numbers that are relatively prime. Problem 1.10. (International Math. Olympiad) Prove that it is impossible to extract an infinite arithmetic progression from the sequence S = {1, 2k , 3k , . . . , nk , . . . }, where k ≥ 2. Problem 1.11. (Amer. Math. Monthly) Prove that ba−j +1 divides Cba if a, b ≥ 2, j ≤ a + 1. j

Problem 1.12. Solve in integers the following equations: (1) x 2 = y 2 + y 3 ; (2) x 2 + y 2 = z2 . Problem 1.13. Let a, b, c, d ∈ Z+ be such that at least one of a and c is not a perfect square and gcd(a, c) = 1. Show that there exist infinitely many natural numbers n such that an + b, cn + d are simultaneously perfect squares if one of the following conditions is satisfied: 1. b and d are perfect squares; 2. a + b, c + d are perfect squares; 3. a(d − 1) = c(b − 1). Moreover, there do not exist such numbers if a = 1, b = 0, c = 4k 2 − 1, d = 1. √ Problem 1.14. If N = 2 + 2 28n2 + 1 ∈ Z for a natural number n, then N is a perfect square. Problem 1.15. Let n ≥ 5, 2 ≤ b ≤ n. Prove that

(n − 1)! ≡ 0 (mod b − 1). b

1 Number Theory

5

Problem 1.16. (Amer. Math. Monthly) Prove that for every natural number n, there exists a natural number k such that k appears in exactly n nontrivial Pythagorean triples. Problem 1.17. (Amer. Math. Monthly) Let n, q ∈ Z+ be such that all prime divisors of q are greater than n. Show that (q − 1)(q 2 − 1) · · · (q n−1 − 1) ≡ 0 (mod n!). Problem 1.18. Every natural number n ≥ 6 can be written as a sum of distinct primes. Problem 1.19. Every number n ≥ 6 can be written as a sum of three numbers that are pairwise relatively prime. Problem 1.20. (Romanian training camp, Sinaia 1984) Find all pairs of integers (m, n) such that n Cm = 1984, n = where Cm

m! n!(m−n)!

denotes the usual binomial coefficient.

Problem 1.21. (Romanian training camp, Sinaia 1984) Find the set A consisting of natural numbers n that are divisible by all odd natural numbers a with a 2 < n. Problem 1.22. Prove that 21092 − 1 is divisible by 10932 . Problem 1.23. (Amer. Math. Monthly) Let n ≥ 0, r > 1, and 0 < a ≤ r be three integers. Prove that the number n, when written in base r, has precisely ∞

nr −k + ar −1 − nr −k

k=1

digits that are greater than or equal to r − a. Problem 1.24. Find a pair (a, b) of natural numbers satisfying the following properties: 1. ab(a + b) is not divisible by 7. 2. (a + b)7 − a 7 − b7 is divisible by 77 . Problem 1.25. (International Math. Olympiad 1984) Let 0 < a < b < c < d be odd integers such that 1. ad = bc 2. a + d = 2k , b + c = 2m , for some integers k and m. Prove that a = 1. Problem 1.26. (Amer. Math. Monthly) Find those subsets S ⊂ Z+ such that all but finitely many sums of elements from S (possibly with repetitions) are composite numbers.

6

1 Number Theory

Problem 1.27. (Amer. Math. Monthly) Prove that for any natural n > 1, the number 2n − 1 does not divide 3n − 1. Problem 1.28. (Putnam Competition 1964) Let un be the least common multiple of the first n terms of a strictly increasing sequence of positive integers. Prove that ∞ 1 ≤ 2. un n=1

Find a sequence for which equality holds above. Problem 1.29. (Amer. Math. Monthly) Let ϕn (m) = ϕ(ϕn−1 (m)), where ϕ1 (m) = ϕ(m) is the Euler totient function, and set ω(m) the smallest number n such that ϕn (m) = 1. If m < 2α , then prove that ω(m) ≤ α. Problem 1.30. Let f be a polynomial with integer coefficients and N (f ) = card{k ∈ Z; f (k) = ±1}. Prove that N (f ) ≤ 2 + deg f , where deg f denotes the degree of f . Problem 1.31. Prove that every integer can be written as a sum of 5 perfect cubes. Problem 1.32. If n ∈ Z, then the binomial coefficient C2n = number of divisors.

(2n)! (n!)2

has an even

Problem 1.33. (Amer. Math. Monthly) Prove that every n ∈ Z+ can be written in precisely k(n) different ways as a sum of consecutive integers, where k(n) is the number of odd divisors of n greater than 1. Problem 1.34. (Amer. Math. Monthly) Let π2 (x) denote the number of twin primes p with p ≤ x. Recall that p is a twin prime if both p and p + 2 are prime. Show that

π n! π (n − 2)! sin (n + 2) sin n π2 (x) = 2 + 2 n+2 2 n 7≤n≤x

for x > 7. Problem 1.35. (Kvant) 1. Find all solutions of the equation 3x+1 + 100 = 7x−1 . 2 2 2. Find two solutions of the equation 3x + 3x = 2x + 4x , and prove that there are no others. Problem 1.36. (Kvant) Let σ (n) denote the sum of the divisors of n. Prove that there exist infinitely many integers n such that σ (n) > 2n, or even stronger, such that σ (n) > 3n. Prove also that σ (n) < n(1 + log n). Problem 1.37. (American Competition) Let ai be natural numbers such that gcd(ai , aj ) = 1 and the ai are not prime numbers. Show that 1 1 + ··· + < 2. a1 an

1 Number Theory

7

Problem 1.38. Let σ (n) denote the sum of divisors of n. Show that σ (n) = 2k if and only if n is a product of Mersenne primes, i.e., primes of the form 2k − 1, for k ∈ Z. Problem 1.39. (Putnam Competition) Find all integer solutions of the equation |pr − q s | = 1, where p, q are primes and r, s ∈ Z \ {0, 1}. Problem 1.40. Consider an arithmetic progression with ratio between 1 and 2000. Show that the progression does not contain more than 10 consecutive primes. √ Problem 1.41. (Amer. Math. Monthly) Let a1 = 1, an+1 = an + an . Show that an is a perfect square iff n is of the form 2k + k − 2. Problem 1.42. Recall that ϕ(n) denotes the Euler totient function (i.e., the number of natural numbers less than n and prime to n), and that σ (n) is the sum of divisors of n. Show that n is prime iff ϕ(n) divides n − 1 and n + 1 divides σ (n). Problem 1.43. (Amer. Math. Monthly) A number is called ϕ-subadditive if ϕ(n) ≤ ϕ(k) + ϕ(n − k) for all k such that 1 ≤ k ≤ n − 1, and ϕ-superadditive if the reverse inequality holds. Prove that there are infinitely many ϕ-subadditive numbers and infinitely many ϕ-superadditive numbers. Problem 1.44. (Amer. Math. Monthly) Find the positive integers N such that for all n ≥ N, we have ϕ(n) ≤ ϕ(N ). Problem 1.45. A number n is perfect if σ (n) = 2n, where σ (n) denotes the sum of all divisors of n. Prove that the even number n is perfect if and only if n = 2p−1 (2p − 1), where p is a prime number with the property that 2p − 1 is prime. Problem 1.46. (Elemente der Mathematik) A number n is superperfect if σ (σ (n)) = 2n, where σ (k) is the sum of all divisors of k. Prove that the even number n is superperfect if and only if n = 2r , where r is an integer such that 2r+1 − 1 is prime. Problem 1.47. If a, b are rational numbers satisfying tan aπ = b, then b ∈ {−1, 0, 1}. Problem 1.48. (Amer. Math. Monthly) Let An = run + sv n , n ∈ Z+ , where r, s, u, v are integers, u = ±v, and let Pn be the set of prime divisors of An . Then P = ∞ n=0 Pn is infinite. Problem 1.49. Solve in natural numbers the equation x 2 + y 2 + z2 = 2xyz. Problem 1.50. (Amer. Math. Monthly) Find the greatest common divisor of the fol1 , C 3 , C 5 , . . . , C 2n−1 . lowing numbers: C2n 2n 2n 2n Problem 1.51. (Amer. Math. Monthly) Prove that if Sk = ni=1 k gcd(i,n) , then Sk ≡ 0 (mod n).

8

1 Number Theory

Problem 1.52. Prove that for any integer n such that 4 ≤ n ≤ 1000, the equation 1 1 1 4 + + = x y z n has solutions in natural numbers. Problem 1.53. (Nieuw Archief v. Wiskunde) For a natural number n we set S(n) for the set of integers that can be written in the form 1 + g + · · · + g n−1 for some g ∈ Z+ , g ≥ 2. 1. Prove that S(3) ∩ S(4) = ∅. 2. Find S(3) ∩ S(5). Problem 1.54. If k ≥ 202 and n ≥ 2k, then prove that Cnk > nπ(k) , where π(k) denotes the number of prime numbers smaller than k. Problem 1.55. (Amer. Math. Monthly) Let Sm (n) be the sum of the inverses of the integers smaller than m and relatively prime to n. If m > n ≥ 2, then show that Sm (n) is not an integer. Problem 1.56. Solve in integers the following equations: 1. x 4 + y 4 = z2 ; 2. y 3 + 4y = z2 ; 3. x 4 = y 4 + z2 . Problem 1.57. Show that the equation x 3 + y 3 = z3 + w 3 has infinitely many integer solutions. Prove that 1 can be written in infinitely many ways as a sum of three cubes. Problem 1.58. Prove that the equation y 2 = Dx 4 + 1 has no integer solution except x = 0 if D ≡ 0, −1, 3, 8 (mod 16) and there is no factorization D = pq, where p > 1 is odd, gcd(p, q) = 1, and either p ≡ ±1 (mod 16), p ≡ q ± 1 (mod 16), or p ≡ 4q ± 1 (mod 16). Problem 1.59. (Amer. Math. Monthly) Find all numbers that are simultaneously triangular, perfect squares, and pentagonal numbers. Problem 1.60. (Amer. Math. Monthly) Find all inscribable integer-sided quadrilaterals whose areas equal their perimeters. Problem 1.61. (Amer. Math. Monthly) Every rational number a can be written as a sum of the cubes of three rational numbers. Moreover, if a > 0, then the three cubes can be chosen to be positive. Problem 1.62. 1. Every prime number of the form 4m + 1 can be written as the sum of two perfect squares.

1 Number Theory

9

2. Every natural number is the sum of four perfect squares. Problem 1.63. Every natural number can be written as a sum of at most 53 integers to the fourth power. Problem 1.64. Let G(k) denote the minimal integer n such that any positive integer k cank be written as the sum of n positive perfect kth powers. Prove that G(k) ≥ 2 + 3 − 2. 2k Problem 1.65. (Amer. Math. Monthly) Let R(n, k) be the remainder when n is divided by k and k R(n, i). S(n, k) = i=1 2

n) = 1 − π12 . 1. Prove that limn→∞ S(n, n2 2. Consider a sequence of natural numbers (ak ) growing to infinity and such that k limk→∞ ak log = 0. Prove that limk→∞ S(kakk2, k) = 41 . k

Problem 1.66. (Nieuw Archief v. Wiskunde) Let a1 < a2 < a3 < · · · < an < · · · be a sequence of positive integers such that the series ∞ 1 ai i=1

converges. Prove that for any i, there exist infinitely many sets of ai consecutive integers that are not divisible by aj for all j > i. Problem 1.67. (Amer. Math. Monthly) Denote by Cn the claim that there exists a set of n consecutive integers such that no two of them are relatively prime. Prove that Cn is true for every n such that 17 ≤ n ≤ 10000. Problem 1.68. (Schweitzer Competition) Prove that a natural number has more divisors that can be written in the form 3k + 1, for k ∈ Z, than divisors of the form 3m − 1, for m ∈ Z. Problem 1.69. (Amer. Math. Monthly) A number N is called deficient if σ (N ) < 2N and abundant if σ (N) > 2N . 1. Let k be fixed. Are there any sequences of k consecutive abundant numbers? 2. Show that there are infinitely many 5-tuples of consecutive deficient numbers. Problem 1.70. (Amer. Math. Monthly) Does there exist a nonconstant polynomial an2 +bn+c with integer coefficients such that for any natural numbers m all its prime factors pi are congruent to 3 modulo 4? Prove that for any nonconstant polynomial f with integer coefficients and any m ∈ Z there exist a prime number p and a natural number n such that p divides f (n) and p ≡ 1 (mod m).

2 Algebra and Combinatorics

2.1 Algebra p−1 Problem 2.1. Set Sk,p = i=1 i k , for natural numbers p and k. If p ≥ 3 is prime and 1 < k ≤ p − 2, show that Sk,p ≡ 0 (mod p). Problem 2.2. Let P = a0 +· · ·+an x n and Q = b0 +· · ·+bm x m be two polynomials with m ≤ n. Then, deg gcd(P , Q) ≥ 1 if and only if there exist two polynomials K and L, such that deg K ≤ m − 1, deg L ≤ n − 1, and K · P = L · Q. Prove that this is equivalent to the vanishing of the following (n + m) × (n + m) determinant: ⎛ ⎞ a0 a1 a2 · · · am−1 · · · an−1 an 0 0 ··· 0 ⎜ 0 a0 a1 · · · am−2 · · · an−2 an−1 an 0 ··· 0 ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. .. .. .. .. ⎟ ⎜ . . . ⎟ . . . . . . ⎜ ⎟ ⎜ 0 0 0 · · · a0 · · · an−m−1 an−m an−m+1 · · · a n ⎟ ⎜ ⎟ ⎜ 0 0 0 ··· 0 · · · an−m−2 an−m−1 an−m ··· an−1 ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. ⎟ = 0. det ⎜ ... ... ... . . . . . ⎟ ⎜ ⎟ ⎜ 0 0 0 ··· 0 0 a0 a1 a2 ··· am ⎟ ⎜ ⎟ ⎜ b0 b1 · · · bm−1 bm 0 0 0 0 ··· 0 ⎟ ⎜ ⎟ ⎜ 0 b0 · · · bm−2 bm−1 bm 0 0 0 ··· 0 ⎟ ⎜ ⎟ ⎜ . . .. .. .. .. .. .. .. ⎟ ⎝ .. .. . . . . . . . ⎠ 0 0 · · · b0 b1 · · · · · · ··· 0 ··· 0 Problem 2.3. Prove that if P , Q ∈ R[x, y] are relatively prime polynomials, then the system of equations P (x, y) = 0, Q(x, y) = 0, has only finitely many real solutions.

12

2 Algebra and Combinatorics

Problem 2.4. Let a, b, c, d ∈ R[x] be polynomials with real coefficients. Set x x x x p= ac dt, q = ad dt, r = bc dt, s = bd dt. 1

1

1

1

Prove that ps − qr is divisible by (x − 1)4 . Problem 2.5. 1. Find the minimum number of elements that must be deleted from the set {1, . . . , 2005} such that the set of the remaining elements does not contain two elements together with their product. 2. Does there exist, for any k, an arithmetic progression with k terms in the infinite sequence 1 1 1 1, , . . . , , . . . , , . . .? 2 2005 n Problem 2.6. (Amer. Math. Monthly) Consider a set S of n elements and n+1 subsets M1 , . . . , Mn+1 ⊂ S. Show that there exist r, s ≥ 1 and disjoint sets of indices {i1 , . . . , ir } ∩ {j1 , . . . , js } = ∅ such that r

Mik =

k=1

s

Mjk .

k=1

Problem 2.7. Let p be a prime number, and A = {a1 , . . . , ap−1 } ⊂ Z∗+ a set of integers that are not divisible by p. Define the map f : P(A) → {0, 1, . . . , p − 1} by k f ({ai1 , . . . , aik }) = aip (mod p), and f (∅) = 0. p−1

Prove that f is surjective. Problem 2.8. Consider the function Fr = x r sin rA + y r sin rB + zr sin rC, where x, y, z ∈ R, A + B + C = kπ, and r ∈ Z+ . Prove that if F1 (x0 , y0 , z0 ) = F2 (x0 , y0 , z0 ) = 0, then Fr (x0 , y0 , z0 ) = 0, for all r ∈ Z+ . Problem 2.9. Let T (z) ∈ Z[z] be a nonzero polynomial with the property that |T (ui )| ≤ 1 for all values ui that are roots of P (z) = zn − 1. Prove that either T (z) is divisible by P (z), or else there exists some k ∈ Z+ , k ≤ n − 1, such that T (z) ± zk is divisible by P (z). The same result holds when instead of P (z), we consider zn + 1. Problem 2.10. 1. If the map x → x 3 from a group G to itself is an injective group homomorphism, then G is an abelian. 2. If the map x → x 3 from a group G to itself is a surjective group homomorphism, then G is an abelian. 3. Find an abelian group with the property that x → x 4 is an automorphism. 4. What can be said for exponents greater than 4?

2.2 Algebraic Combinatorics

13

Problem 2.11. (Amer. Math. Monthly) Let V be a vector space of dimension n > 0 over a field of characteristic p = 0 and let A be an affine map A : V → V . Prove that there exist u ∈ V and 0 ≤ k ≤ np such that Ak u = u. Problem 2.12. (Putnam Competition 1959) Find the cubic equation the zeros of which are the cubes of the roots of the equation x 3 + ax 2 + bx + c = 0. Problem 2.13. (Putnam Competition 1956) Assume that the polynomials P , Q ∈ C[x] have the same roots, possibly with different multiplicities. Suppose, moreover, that the same holds true for the pair P + 1 and Q + 1. Prove that P = Q. Problem 2.14. (Amer. Math. Monthly) Determine r ∈ Q, for which 1, cos 2π r, sin 2π r are linearly dependent over Q. Problem 2.15. (Amer. Math. Monthly) 1. Prove that there exist a, b, c ∈ Z, not all zero, such that |a|, |b|, |c| < 106 √ √ a + b 2 + c 3 < 10−11 . 2. Prove that if 0 ≤ |a|, |b|, |c| 10−21 . Problem 2.16. (Amer. Math. Monthly) Prove that if n > 2, then we do not have any nontrivial solutions for the equation x n + y n = zn , where x, y, z are rational functions. Solutions of the form x = af, y = bf, z = cf , where f is a rational function and a, b, c are complex numbers satisfying a n +bn = cn , are called trivial. Problem 2.17. (Kvant) A table is an n × k rectangular grid drawn on the torus, every box being assigned an element from Z/2Z. We define a transformation acting on tables as follows. We replace all elements of the grid simultaneously, each element being changed into the sum of the numbers previously assigned to its neighboring boxes. Prove that iterating this transformation sufficiently many times, we always obtain the trivial table filled with zeros, no matter what the initial table was, if and only if n = 2p and k = 2q for some integers p, q. In this case we say that the respective n × k grid is nilpotent.

2.2 Algebraic Combinatorics Problem 2.18. Let us consider a four-digit number N whose digits are not all equal. We first arrange its digits in increasing order, then in decreasing order, and finally, we subtract the two obtained numbers. Let T (N) denote the positive difference thus obtained. Show that after finitely many iterations of the transformation T , we obtain 6174.

14

2 Algebra and Combinatorics

Problem 2.19. Find an example of a sequence of natural numbers 1 ≤ a1 < a2 < · · · < an < an+1 < · · · with the property that every m ∈ Z+ can be uniquely written as m = ai − aj , for i, j ∈ Z+ . Problem 2.20. (Amer. Math. Monthly) Consider the set of 2n integers {±a1 , ±a2 , . . . , ±an } and m < 2n . Show that we can choose a subset S such that 1. The two numbers ±ai are not both in S; 2. The sum of all elements of S is divisible by m. Problem 2.21. (Proposed for the International Math. Olympiad) Show that for every natural number n there exist prime numbers p and q such that n divides their difference. Problem 2.22. An even number, 2n, of knights arrive at King Arthur’s court, each one of them having at most n − 1 enemies. Prove that Merlin the wizard can assign places for them at a round table in such a way that every knight is sitting only next to friends. Problem 2.23. (Putnam Competition) Let r, s ∈ Z+ . Find the number of 4-tuples of positive integers (a, b, c, d) that satisfy 3r 7s = lcm(a, b, c) = lcm(a, b, d) = lcm(a, c, d) = lcm(b, c, d). Problem 2.24. (Putnam Competition) 1. Let n ∈ Z+ and p be a prime number. Denote by N (n, p) the number of binomial coefficients Cns that are not divisible by p. Assume that n is written in base p as n = n0 + n1 p + · · · + nm p m , where 0 ≤ nj < p for all j ∈ {0, 1, . . . , m}. Prove that N (n, p) = (n0 + 1)(n1 + 1) · · · (nj + 1). 2. Write k in base p as k = k0 + k1 p + · · · + kj p j , with 0 ≤ kj ≤ p − 1, for all j ∈ {0, 1, . . . , s}. Prove that k

Cnk ≡ Cnk00 Cnk11 · · · Cnjj (mod p). Problem 2.25. (Putnam Competition) Define the sequence Tn by T1 = 2, Tn+1 = Tn2 − Tn + 1, for n ≥ 1. Prove that if m = n, then Tm and Tn are relatively prime, and further, that ∞ 1 = 1. Ti i=1

Problem 2.26. Let α, β > 0 and consider the sequences [α], [2α], . . . , [kα], . . . ; [β], [2β], . . . , [kβ], . . . , where the brackets denote the integer part. Prove that these two sequences taken together enumerate Z+ in an injective manner if and only if α, β ∈ R \ Q and

1 1 + = 1. α β

2.2 Algebraic Combinatorics

15

Problem 2.27. We say that the sets S1 , S2 , . . . , Sm form a complementary system if they make a partition of Z+ , i.e., every positive integer belongs to a unique set Si . Let m > 1 and α1 , . . . , αm ∈ R+ . Then the sets Si = {[nαi ], where n ∈ Z+ } form a complementary system only if m = 2, α1−1 + α2−1 = 1, and α1 ∈ R \ Q. Problem 2.28. Let f : Z+ → Z+ be an increasing function and set F (n) = f (n) + n, G(n) = f ∗ (n) + n, where f ∗ (n) = card({x ∈ Z+ ; 0 ≤ f (x) < n}). Then {F (n); n ∈ Z+ } and {G(n); n ∈ Z+ } are complementary sequences. Conversely, any two complementary sequences can be obtained this way using some nondecreasing function f . Problem 2.29. Let M denote the set of bijective functions f : Z+ → Z+ . Prove that there is no bijective function between M and Z. Problem 2.30. (Amer. Math. Monthly) Let F ⊂ Z be a finite set of integers satisfying the following properties: 1. For any x ∈ F there exist y, z ∈ F such that x = y + z. 2. There exists n such that for any natural number 1 ≤ k ≤ n, and any choice of x1 , . . . , xk ∈ F , their sum x1 + · · · + xk is nonzero. Prove that card(F ) ≥ 2n + 2. Problem 2.31. (Amer. Math. Monthly) For a finite graph G we denote by Z(G) the minimal number of colors needed to color all its vertices such that adjacent vertices have different colors. This is also called the chromatic number of G. Prove that the inequality Z(G) ≥

p2

p2 − 2q

holds if G has p vertices and q edges. Problem 2.32. Let Dk be a collection of subsets of the set {1, . . . , n} with the property that whenever A = B ∈ Dk , then card(A ∩ B) ≤ k, where 0 ≤ k ≤ n − 1. Prove that card(Dk ) ≤ Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . Problem 2.33. Prove that

⎧ 0, ⎪ ⎪ ⎪ ⎪ 1, ⎪ ⎪ ⎪ ⎨ n/2,

if if n if 1 (−1)n−k Cnk k p = n(3n+1) , if ⎪ p! 24 ⎪ k=0 ⎪ n2 (n+1) ⎪ ⎪ , if ⎪ 48 ⎪ ⎩ n(15n3 +30n2 +5n+1) , if 1152

0 ≤ p < n, p = n, p = n + 1, p = n + 2, p = n + 3, p = n + 4.

16

2 Algebra and Combinatorics

Problem 2.34. Write lcm(a1 , . . . , an ) in terms of the various gcd(ai , . . . , aj ) for subsets of {a1 , . . . , an }. Problem 2.35. Let f (n) be the number of ways in which a convex polygon with n+1 sides can be divided into regions delimited by several diagonals that do not intersect (except possibly at their endpoints). We consider as distinct the dissection in which we first cut the diagonal a and next the diagonal b from the dissection in which we first cut the diagonal b and next the diagonal a. It is easy to compute the first values of f (n), as follows: f (1) = 1, f (2) = 1, f (3) = 3, f (4) = 11, f (5) = 45. Find the generating function F (x) = f (n)x n and an asymptotic formula for f (n). Problem 2.36. Find the permutation σ : (1, . . . , n) → (1, . . . , n) such that S(σ ) =

n

|σ (i) − i|

i=1

is maximal. Problem 2.37. (Amer. Math. Monthly) On the set Sn of permutations of {1, . . . , n} we define an invariant distance function by means of the formula d(σ, τ ) =

n

|σ (i) − τ (i)|.

i=1

What are the values that d could possibly take? Problem 2.38. (Preliminaries for the International Math. Olympiad, Romania) The set M = {1, 2, . . . , 2n} is partitioned into k sets M1 , . . . , Mk , where n ≥ k 3 + k. Show that there exist i, j ∈ {1, . . . , k} for which we can find k + 1 distinct even numbers 2r1 , . . . , 2rk+1 ∈ Mi with the property that 2r1 − 1, . . . , 2rk+1 − 1 ∈ Mj . Problem 2.39. (Preliminaries for the International Math. Olympiad, Great Britain) Let S be the set of odd integers not divisible by 5 and smaller than 30m, where m ∈ Z∗+ . Find the smallest k such that every subset A ⊂ S of k elements contains two distinct integers, one of which divides the other. ai −aj Problem 2.40. Prove that 1≤j nn−1 . Show that there exist distinct prime numbers p1 , p2 , . . . , pn such that C + j is divisible by pj . Problem 2.47. (Nieuw Archief v. Wiskunde) Let p be a prime number and f (p) the smallest integer for which there exists a partition of the set {2, 3, . . . , p} into f (p) classes such that whenever a1 , . . . , ak belong to the same class of the partition, the equation k xi ai = p i=1

does not have solutions in nonnegative integers. Estimate f (p). Problem 2.48. (Schweitzer Competition) Consider m distinct natural numbers ai √ smaller than N such that lcm(ai , aj ) ≤ N for all i, j . Prove that m ≤ 2 N . Problem 2.49. The set M ⊂ Z+ is called A-sum-free, where A = (a1 , a2 , . . . , ak ) ∈ Zk+ , if for any choice of x1 , x2 , . . . , xk ∈ M we have a1 x1 + a2 x2 + · · · + ak xk ∈ M. If A, B are two vectors, we define f (n; A, B) as the greatest number h such that there exists a partition of the set of consecutive integers {n, n + 1, . . . , h} into S1 and S2 such that S1 is A-sum-free and S2 is B-sum-free. Assume that B = (b1 , b2 , . . . , bm ) and that the conditions below are satisfied: a1 + a2 + · · · + ak = b1 + b2 + · · · + bm = s, and min aj = min bj = 1, k, m ≥ 2.

1≤j ≤k

Prove that f (n; A, B) =

ns 2

1≤j ≤m

+ n(s − 1) − 1.

Problem 2.50. (Amer. Math. Monthly) Let 1 ≤ a1 < a2 < · · · < an < 2n be a sequence of natural numbers for n ≥ 6. Prove that n min lcm(ai , aj ) ≤ 6 +1 . i,j 2 Moreover, the constant 6 is sharp.

18

2 Algebra and Combinatorics

Problem 2.51. (Amer. Math. Monthly) Let 1 ≤ a1 < a2 < · · · < ak < n be such that gcd(ai , aj ) = 1 for all 1 ≤ i < j ≤ k. Determine the maximum value of k. Problem 2.52. (International Math. Olympiad 1978) Consider the increasing sequence f (n) ∈ Z+ , 0 < f (1) < f (2) < · · · < f (n) < · · · . It is known that the nth element in increasing order among the positive integers that are not terms of this sequence is f (f (n)) + 1. Find the value of f (240). Problem 2.53. (Amer. Math. Monthly) We define inductively three sequences of integers (an ), (bn ), (cn ) as follows: 1. a1 = 1, b1 = 2, c1 = 4; 2. an is the smallest integer that does not belong to the set {a1 , . . . , an−1 , b1 , . . . , bn−1 , c1 , . . . , cn−1 }; 3. bn is the smallest integer that does not belong to the set {a1 , . . . , an−1 , an , b1 , . . . , bn−1 , c1 , . . . , cn−1 }; 4. cn = 2bn + n − an . Prove that

√ 0 < n 1 + 3 − bn < 2

for all n ∈ Z+ .

2.3 Geometric Combinatorics Problem 2.54. We consider n points in the plane that determine Cn2 segments, and to each segment one associates either +1 or −1. A triangle whose vertices are among these points will be called negative if the product of numbers associated to its sides is negative. Show that if n is even, then the number of negative triangles is even. Moreover, for odd n, the number of negative triangles has the same parity as the number p of segments labeled −1. Problem 2.55. (Amer. Math. Monthly) Given n find a finite set S consisting of natural numbers larger than n with the property that for any k ≥ n the k × k square can be tiled by a family of si × si squares, where si ∈ S. Problem 2.56. We consider 3n points A1 , . . . , A3n in the plane whose positions are defined recursively by means of the following rule: first, the triangle A1 A2 A3 is equilateral; further, the points A3k+1 , A3k+2 , and A3k+3 are the midpoints of the sides of the triangle A3k A3k−1 A3k−2 . Let us assume that the 3n points are colored with two colors. Show that for n ≥ 7 there exists at least one isosceles trapezoid having vertices of the same color. Problem 2.57. Is there a coloring of all lattice points in the plane using only two colors such that there are no! rectangles with all vertices of the same color whose side ratio belongs to 1, 21 , 13 , 23 ?

2.3 Geometric Combinatorics

19

Problem 2.58. (Amer. Math. Monthly) Let G be a planar graph and let P be a path in G. We say that P has a (transversal) self-intersection in the vertex v if the path has a (transversal) self-intersection from the curve-theoretic viewpoint. Let us give an example: Take the point 0 in the plane and the segments 01, 02, 03, 04 going counterclockwise around 0. Then a path traversing first 103 and then 204 has a (transversal) self-intersection at 0, while a path going first along 102 and further on 304 does not have a (transversal) self-intersection. Prove that any connected planar graph G with only even-degree vertices admits an Eulerian circuit without self-intersections. Recall that an Eulerian circuit is a path along the edges of the graph that passes precisely once along each edge of the graph. Problem 2.59. (Hungarian Competition) Let us consider finitely many points in the plane that are not all collinear. Assume that one associates to each point a number from the set {−1, 0, 1} such that the following property holds: for any line determined by two points from the set, the sum of numbers associated to all points lying on that line equals zero. Show that, if the number of points is at least three, then to each point one is associated with 0. Problem 2.60. (Amer. Math. Monthly) If one has a set of squares with √ total area smaller than 1, then one can arrange them inside a square of side length 2, without any overlaps. Problem 2.61. Prove that for each k there exist k points in the plane, no three collinear and having integral distance from each other. If we have an infinite set of points with integral distances from each other, then all points are collinear. Problem 2.62. (International Math. Olympiad 1984) Let O, A be distinct points in (counterclockwise). the plane. For each point x in the plane, we write α(x) = xOA α(x) Let C(x) be the circle of center O and radius |Ox| + |Ox| . If the points in the plane are colored with finitely many colors, then there exists a point y with α(y) > 0 such that the color of y also belongs to the circle C(y). Problem 2.63. (Amer. Math. Monthly) Let k, n ∈ Z+ . 1. Assume that n − 1 ≤ k ≤ n(n−1) 2 . Show that there exist n distinct points x1 , ..., xn on a line that determine exactly k distinct distances |xi − xj |. 2. Suppose that [ n2 ] ≤ k ≤ n(n−1) 2 . Then there exist n points in the plane that determine exactly k distinct distances. 3. Prove that for any ε > 0, there exists some constant n0 = n0 (ε) such that for any n > n0 and εn < k < n(n−1) 2 , there exist n points in the plane that determine exactly k distinct distances. Problem 2.64. (International Math. Olympiad 1978) Show that it is possible to pack 2n(2n + 1) nonoverlapping pieces having the form of a parallelepiped of dimensions 1 × 2 × (n + 1) in a cubic box of side 2n + 1 if and only if n is even or n = 1.

20

2 Algebra and Combinatorics

Problem 2.65. (Putnam Competition 1964) Let F be a finite subset of R with the property that any value of the distance between two points from F (except for the largest one) is attained at least twice, i.e., for two distinct pairs of points. Prove that the ratio of any two distances between points of F is a rational number.

3 Geometry

3.1 Synthetic Geometry Problem 3.1. Let I be the center of the circle inscribed in the triangle ABC and consider the points α, β, γ situated on the perpendiculars from I on the sides of the triangle ABC such that |I α| = |Iβ| = |I γ |. Prove that the lines Aα, Bβ, Cγ are concurrent. Problem 3.2. We consider the angle xOy and a point A ∈ Ox. Let (C) be an arbitrary circle that is tangent to Ox and Oy at the points H and D, respectively. Set AE for the tangent line drawn from A to the circle (C) that is different from AH . Show that the line DE passes through a fixed point that is independent of the circle (C) chosen above. Problem 3.3. Let C be a circle of center O and A a fixed point in the plane. For any with point P ∈ C, let M denote the intersection of the bisector of the angle AOP the circle circumscribed about the triangle AOP . Find the geometric locus of M as P runs over the circle C. Problem 3.4. Let ABC be an isosceles triangle having |AB| = |AC|. If AS is an interior Cevian that intersects the circle circumscribed about ABC at S, then describe the geometric locus of the center of the circle circumscribed about the triangle BST , where {T } = AS ∩ BC. Problem 3.5. Let AB, CD, EF be three chords of length one on the unit circle. Then the midpoints of the segments |BC|, |DE|, and |AF | form an equilateral triangle. Problem 3.6. (Amer. Math. Monthly) Denote by P the set of points of the plane. Let : P × P → P be the following binary operation: A B = C, where C is the unique point in the plane such that ABC is an oriented equilateral triangle whose orientation is counterclockwise. Show that is a nonassociative and noncommutative operation satisfying the following “medial property”:

22

3 Geometry

(A B) (C D) = (A C) (B D). Problem 3.7. Consider two distinct circles C1 and C2 with nonempty intersection and let A be a point of intersection. Let P , R ∈ C1 and Q, S ∈ C2 be such that P Q and RS are the two common tangents. Let U and V denote the midpoints of the chords P R and QS. Prove that the triangle AU V is isosceles. Problem 3.8. (Amer. Math. Monthly) If the planar triangles AU V , V BU , and U V C are directly similar to a given triangle, then so is ABC. Recall that two triangles are directly similar if one can obtain one from the other using a homothety with positive ratio, rotations and translations. Problem 3.9. (Amer. Math. Monthly) Find, using a straightedge and a compass, the directrix and the focus of a parabola. Recall that the parabola is the geometric locus of those points P in the plane that are at equal distance from a point O (called the focus) and a line d called the directrix. Problem 3.10. Prove that if M is a point in the interior of a circle and AB ⊥ CD are two chords perpendicular at M, then it is possible to construct an inscribable quadrilateral with the following lengths: |AM| − |MB|, |AM| + |MB|, |DM| − |MC|, |DM| + |MC|. Problem 3.11. (Amer. Math. Monthly) If the Euler line of a triangle passes through the Fermat point, then the triangle is isosceles. Problem 3.12. Consider a point M in the interior of the triangle ABC, and choose A ∈ AM, B ∈ BM, and C ∈ CM. Let P , Q, R, S, T , and U be the intersections of the sides of ABC and A B C . Show that P S, T Q, and RU meet at M. Problem 3.13. Show that if an altitude in a tetrahedron crosses two other altitudes, then all four altitudes are concurrent. Problem 3.14. Three concurrent Cevians in the interior of the triangle ABC meet the corresponding opposite sides at A1 , B1 , C1 . Show that their common intersection point is uniquely determined if |BA1 |, |CB1 |, and |AC1 | are equal. Problem 3.15. (International Math. Olympiad 1983) Let ABCD be a convex quadrilateral with the property that the circle of diameter AB is tangent to the line CD. Prove that the circle of diameter CD is tangent to the line AB if and only if AD is parallel to BC. Problem 3.16. Let A , B , C be points on the sides BC, CA, AB of the triangle ABC. Let M1 , M2 be the intersections of the circle A B C with the circle ABA and let N1 , N2 be the analogous intersections of the circle A B C with the circle ABB . 1. Prove that M1 M2 , N1 N2 , AB are either parallel or concurrent, in a point that we denote by A1 ;

3.2 Combinatorial Geometry

23

2. Prove that the analogously defined points A1 , B1 , C1 are collinear. Problem 3.17. (Putnam Competition) A circumscribable quadrilateral of area S = √ abcd is inscribable. Problem 3.18. (Nieuw Archief v. Wiskunde) Let O be the center of the circumcircle, Ge the Gergonne point, N a the Nagel point, and G1 , N1 , the isogonal conjugates of G and N, respectively. Prove that G1 , N1 , and O are collinear (see also the Glossary for definitions of the important points in a triangle).

3.2 Combinatorial Geometry Problem 3.19. Consider a rectangular sheet of paper. Prove that given any ε > 0, one can use finitely many foldings of the paper along its sides in either 2 equal parts or 3 equal parts to obtain a rectangle whose sides are in ratio r for some r satisfying 1 − ≤ r ≤ 1 + . Problem 3.20. (Amer. Math. Monthly) Show that there exist at most√three points on the unit circle with the distance between any two being greater than 2. Problem 3.21. (Komal) A convex polygon with 2n sides has at least n diagonals not parallel to any of its sides. Problem 3.22. Let d be the sum of the lengths of the diagonals of a convex polygon P1 . . . Pn and p its perimeter. Prove that for n ≥ 4, we have n n + 1 d n−3 10−21 , because 1/|Fi | > 10−7 . Problem 2.16. Prove that if n > 2, then we do not have any nontrivial solutions for the equation x n + y n = zn , where x, y, z are rational functions. Solutions of the form x = af, y = bf, z = cf , where f is a rational function and a, b, c are complex numbers satisfying a n +bn = cn , are called trivial.

6.1 Algebra

93

Solution 2.16. Any rational solution yields a polynomial solution, by clearing the denominators. Let further (f, g, h) be a polynomial solution for which the degree r = max{deg(f ), deg(g), deg(h)} is minimal among all polynomial solutions, and where r > 0. Consider the root of unity ξ = exp 2πi n and assume that f, g, and h are relatively prime. The equation can be written as n−1 &

(f − ξ j h) = g n .

j =0

Now, gcd(f, h) and gcd(f, g) are relatively prime, so that f − ξ j h and f − ξ k h are relatively prime when k = j . Since the factorization of polynomials is unique, we must have g = g1 · · · gn−1 , where gj are polynomials satisfying gjn = f − ξ j h. Consider now the set {f − h, f − ξ h, f − ξ 2 h}. Since n > 2, these elements belong to the two-dimensional space generated by f and h over C. Thus there exists a vanishing linear combination with complex coefficients in these three elements. √ Thus, there exist ai ∈ C so that a0 g0n + a1 g1n = a2 g2n . We then set hj = n aj gj , and observe that hn0 + hn1 = hn2 . Moreover, the polynomials hi and hj are relatively prime if i = j and max deg(hi ) < r, which contradicts our choice of r. This proves the claim. Problem 2.17. A table is an n × k rectangular grid drawn on the torus, every box being assigned an element from Z/2Z. We define a transformation acting on tables as follows. We replace all elements of the grid simultaneously, each element being changed into the sum of the numbers previously assigned to its neighboring boxes. Prove that iterating this transformation sufficiently many times, we always obtain the trivial table filled with zeros, no matter what the initial table was, if and only if n = 2p and k = 2q , for some integers p, q. In this case, we say that the respective n × k grid is nilpotent. Solution 2.17. Consider the m × m square matrices (m ≥ 2) ⎛ ⎞ 0 1 0 0 ··· 0 1 ⎜1 0 1 0 ··· 0 0⎟ ⎜ ⎟ ⎜0 1 0 1 ··· 0 0⎟ ⎜ ⎟ Dm = ⎜ . . . . . . . ⎟ . ⎜ .. .. .. .. .. .. .. ⎟ ⎜ ⎟ ⎝0 0 0 0 ··· 0 1⎠ 1 0 0 0 ··· 1 0 We will write ≡ below if the equality holds (only) for matrices of entries modulo 2. Then the transformation T from the statement acts on the vector space Mn,k (Z/2Z) of n × k matrices with entries in Z/2Z as follows: T (X) = Dn X + XDk , for X ∈ Mn,k (Z/2Z).

94

6 Algebra and Combinatorics Solutions

By induction on s, we have that s

s

s

T 2 (X) ≡ Dn2 X + XDk2 . Thus there exists N such that T N (X) ≡ 0 if and only if there exists some s such that s s Dn2 X ≡ XDk2 . Moreover, we can prove easily that for two matrices A and B of appropriate sizes, we have AX ≡ XB for all X ∈ Mn,k (Z/2Z) if and only if A and B are multiples of the identity by the same scalar. This follows, for instance, if we take X having all but one column (or line) trivial. This assertion is a special case of Schur’s lemma. s Thus T is nilpotent if and only if there exist s and β ∈ Z/2Z such that Dn2 = β ·1n s and Dk2 = β · 1k . Thus the n × k grid is nilpotent if and only if the n × 1 and 1 × k grids are nilpotent. Let us analyze the case of the n × 1 grid. Set a = (a1 , . . . , an ) and a[k] = Dnk a = [k] a1 , . . . , an[k] . By induction, we have k

as[2 ] ≡ as+2k−1 + as−2k−1 , q

where the indices are modulo n. In particular, if n = 2q , then Dn2 = 0 and the grid is nilpotent. This proves the “if” part of the statement. Assume next that n is odd and there exists some r such that a[r] ≡ 0, for any a. Take such an r that is minimal. If n ≥ 3, then r ≥ 3. We claim now that as[r−1] ≡ 1 for [r−1] [r−1] all s. This is true at least for one s, since r is minimal. Since as[r] ≡ as+1 + as−1 , we find that the terms corresponding to even indices (modulo n) are all equal. But n [r−2] [r−2] + as+1 . We is odd and thus all terms are equal. Further, we have as[r−1] = as−1 obtain then that n≡

n

as[r−1] =

s=1

n

n [r−2] [r−2] + as+1 as[r−2] ≡ 0 (mod 2), as−1 =2

s=1

s=1

which is a contradiction. This proves actually that a[r] ≡ 0 when n is odd if and only if a ∈ {(0, 0, . . . , 0), (1, 1, . . . , 1)}. Finally, consider n = 2m h, with h odd. For a vector a of length n, we set s (a) = (as , as+2m , . . . , as+(h−1)2m ). We observe now that

m

s (T 2 (a)) = T (s (a)) m+1

because as[2 ] = as + as+2m . Since the vector s (a) is of length h (which is odd), the previous claim shows us that T N (s (a)) ≡ 0 for some N only if s (a) ∈ {(0, 0, . . . , 0), (1, 1, . . . , 1)}. Moreover, this happens for all values of s. This condition is trivial only when h = 1. This proves the “only if” part. Comments 37 This result was stated in the 1980s as an open question in the journal Kvant. The same method can be used to show that the n × k grids in the plane are

6.2 Algebraic Combinatorics

95

nilpotent if and only if n = 2p − 1 and k = 2q − 1 for some natural numbers p, q. A nice corollary is the following. Define the sequence (αk ) by the recurrence relations α1 = 0, αk+1 = min(2αk + 2, 2N − 2αk − 1). Then there exists some k such that αk = N − 1 if and only if the parameter N has the form 2q − 1 for some natural number q. The solution given here is from: •

V. Boju, L. Funar: Iterative processes for Zn2 , Analele Univ. Craiova 15 (1987), 33–38.

6.2 Algebraic Combinatorics Problem 2.18. Let us consider a four-digit number N whose digits are not all equal. We first arrange its digits in increasing order, then in decreasing order, and finally, we subtract the two obtained numbers. Let T (N) denote the positive difference thus obtained. Show that after finitely many iterations of the transformation T , we obtain 6174. Solution 2.18. Let N have the digits a, b, c, d. 1. Assume that a ≥ b = c ≥ d. Then in T (N), the sum of the first and the fourth digits is 9, as well as the sum of the second and the third. These give five combinations of four digits that give 6174, by applying T once more. 2. Otherwise, a ≥ b > c ≥ d. Then in T (N), the sum of the digits placed at extremities is 10, while the sum of the middle ones is 8. This gives 25 combinations all leading to the desired result, namely 6174. Comments 38 Let a be a positive integer having r digits, not all equal, in base g. Let a be the g-adic integer formed by arranging the digits of a in descending order and let a be that formed by ascending order. Define T (a) = a − a . Then a is said to be self-producing if T (a) = a. There exists an algorithm for obtaining self-producing numbers in any base. Such a fixed point k of T is called a (g-adic) Kaprekar constant if it has the further property that every r-digit integer a eventually yields k on repeated iteration of T and moreover, k is fixed by T . D.R. Kaprekar found that 6174 has this nice property in a short note published in 1949. Ludington proved that for large r ≥ ng there does not exist a Kaprekar constant. Here ng is given by ⎧ if g = 2k, ⎨ g − 1 + (k − 1)δ, ng = 2k + 1 + (g − k − 2)δ, if g = 2t k + 2t−1 + 1, ⎩ 3k − 4 + (g − 1)δ, if g = 2t + 1, and δ ∈ {0, 1} is equal to g (mod 2). Further improvements by Lapenta, Ludington, and Prichett showed that there is no Kaprekar constant in base 10 when the number of digits is r > 4.

96

6 Algebra and Combinatorics Solutions

Hasse and Prichett found that there exists a 4-digit Kaprekar constant in base g if and only if g = 2n · 5, where n is either 0 or an odd number. Moreover, the situation is completely understood for If the base g 3-digit numbers. r is odd, then there is a Kaprekar constant given by r−2 , r − 1, , which is reached 2 2 r+2 within 2 iterations of T (respectively 1, if r = 2). For instance, 495 is a Kaprekar constant if g = 10. If the base g is even, then there is no Kaprekar constant. More precisely, iterations of T willeventually reach the 2 formed by the pair r−1 loop of length r+1 r−1 r+1 of numbers r−3 , , after at most , r − 1, , r − 1, 2 2 2 2 2 steps (respectively 1 if r = 3). •

• • • •

K.E. Eldridge and S. Sagong: The determination of Kaprekar convergence and loop convergence of all three-digit numbers, Amer. Math. Monthly 95 (1988), 105–112. J.F. Lapenta, A.L. Ludington, and G.D. Prichett: An algorithm to determine selfproducing r-digit g-adic integers, J. Reine Angew. Math. 310 (1979), 100–110. G.D. Prichett, A.L. Ludington, and J.F. Lapenta: The determination of all decadic Kaprekar constants, Fibonacci Quart. 19 (1981), 45–52. A.L. Ludington: A bound on Kaprekar constants, J. Reine Angew. Math. 310 (1979), 196–203. H. Hasse and G.D. Prichett: The determination of all four-digit Kaprekar constants, J. Reine Angew. Math. 299/300 (1978), 113–124.

Problem 2.19. Find an example of a sequence of natural numbers 1 ≤ a1 < a2 < · · · < an < an+1 < · · · with the property that every m ∈ Z+ can be uniquely written as m = ai − aj , for i, j ∈ Z+ . Solution 2.19. We consider the sequence a1 = 1, a2 = 2, a2n+1 = 2a2n , a2n+2 = a2n+1 + rn , where rn is the smallest natural number that cannot be written in the form ai − aj , with i, j ≤ 2n + 1. Comments 39 One does not know the minimal growth of such a sequence ak . Problem 2.20. Consider the set of 2n integers {±a1 , ±a2 , . . . , ±an } and m < 2n . Show that we can choose a subset S such that 1. The two numbers ±ai are not both in S; 2. The sum of all elements of S is divisible by m. Solution 2.20. Let S1 , . . . , S2n −1 be the 2n − 1 nonempty distinct subsets of the set {a1 , . . . , an }, where ai ≥ 0. Let F (Si ) denote the sum of the elements of Si . Then, by the pigeonhole principle, there exist i, j such that F (Si ) ≡ F (Sj ) (mod m). Consider next the set S = {Si \ Sj } ∪ −{Sj \ Si }. We derive that F (S) ≡ 0 (mod m).

6.2 Algebraic Combinatorics

97

Problem 2.21. Show that for every natural number n there exist prime numbers p and q such that n divides their difference. Solution 2.21. Consider the following arithmetic progression: 1, 1+n, 1+2n, . . . , 1+ rn, . . . . According to Dirichlet’s theorem, there exist infinitely many prime numbers among the terms of this progression. Let p, q be two of these prime numbers; then p = 1 + nr and q = 1 + ns, where r = s. This yields n(r − s) = p − q, as claimed. An alternative solution is to consider the set of n arithmetic progressions of ratio n starting respectively at 0, 1, 2, . . . , n − 1. Since the set of prime numbers is infinite (this is elementary), there exists at least one progression having infinitely many prime numbers among its terms. The argument above settles the claim. Problem 2.22. An even number, 2n, of knights arrive at King Arthur’s court, each one of them having at most n − 1 enemies. Prove that Merlin the wizard can assign places for them at a round table in such a way that every knight is sitting only next to friends. Solution 2.22. 1. We consider the friendship graph G defined below: its vertices are in bijection with the knights and the edges are joining pairs of vertices whose respective knights are not enemies. The degree of each vertex is at least n. According to Dirac’s theorem, such a graph admits a Hamiltonian cycle, and this yields the wanted assignment of places. 2. Choose an arbitrary assignment of places in which we have two neighbors, A and B, who are enemies. Let us assume that A lies on the right-hand side of B. ˜ B, ˜ According to the pigeonhole principle, there exists another pair of enemies, say A, ˜ who are neighbors, and moreover, A˜ is on the right of B. Let us switch all the places starting at A (and lying on the right side of A) and ˜ using a symmetry. After such a transformation, the number of enemy ending at B, pairs (A, B) is diminished by 2. Applying such transforms iteratively, one obtains a position in which all neighbors are friends. Problem 2.23. Let r, s ∈ Z+ . Find the number of 4-tuples of positive integers (a, b, c, d) that satisfy 3r 7s = lcm(a, b, c) = lcm(a, b, d) = lcm(a, c, d) = lcm(b, c, d). Solution 2.23. The numbers a, b, c, d are of the form 3mi 7ni , 1 ≤ i ≤ 4, where 0 ≤ mi ≤ r and 0 ≤ ni ≤ s. Also, mi = r for at least two values of i, and ni = s for at least two values of i. We have then: (1) one possibility that mi = r for all four i; (2) 4r possibilities that precisely one mi ∈ {0, . . . , r − 1}; that exactly (3) C42 r 2 possibilities two mi ∈ {0, . . . , r − 1}. Therefore, there are 1 + 4r + 6r 2 possibilities for the mi ’s and a similar number for the ni ’s, yielding a total of 1 + 4r + 6r 2 1 + 4s + 6s 2 possibilities. Problem 2.24. 1. Let n ∈ Z+ and p be a prime number. Denote by N (n, p) the number of binomial coefficients Cns that are not divisible by p. Assume that n is written in base p as n = n0 + n1 p + · · · + nm p m , where 0 ≤ nj < p, for all j ∈ {0, 1, . . . , m}. Prove that N (n, p) = (n0 + 1)(n1 + 1) · · · (nm + 1).

98

6 Algebra and Combinatorics Solutions

2. Write k in base p as k = k0 + k1 p + · · · + ks p s , with 0 ≤ kj ≤ p − 1, for all j ∈ {0, 1, . . . , s}. Prove that Cnk ≡ Cnk00 Cnk11 · · · Cnkss (mod p). Solution 2.24. 1. It is clear that Cpk ≡ 0 (mod p) for all k ∈ {1, 2, . . . , p}. Thus n n (1 + x)p ≡ 1 + x p in (Z/pZ)[x]. By induction, we find that (1 + x)p ≡ 1 + x p in (Z/pZ)[x] for any natural number n. We have then the following congruences in (Z/pZ)[x]: m

m

(1+x)n = (1+x)n0 (1+x)n1 p · · · (1+x)nm p ≡ (1+x)n0 (1+x p )n1 · · · (1+x p )nm . When developing factors on the right-hand side, we obtain a sum of factors 2 m x a0 +a1 p+a2 p +···+am p , with a coefficient that is nonzero modulo p. Since every number can be uniquely written in base p, all these factors are distinct and their coefficients modulo p are nonzero. There are exactly (n0 + 1)(n1 + 1) · · · (nm + 1) such factors, and therefore as many binomials not divisible by p. 2 m 2. As observed above, the factor x k appears in the form x a0 +a1 p+a2 p +···+am p ; since k can be uniquely written in base p, we have ai = ki . This implies that the k coefficient of x k is Cnk00 Cnk11 · · · Cnjj , and hence the claim. Problem 2.25. Define the sequence Tn by T1 = 2, Tn+1 = Tn2 − Tn + 1, for n ≥ 1. Prove that if m = n, then Tm and Tn are relatively prime, and further, that ∞ 1 = 1. Ti i=1

Solution 2.25. We prove by induction that Tn+1 = 1 + T1 T2 · · · Tn . In fact, we have Tn+1 = Tn2 − Tn + 1 = Tn (Tn − 1) + 1 = Tn (Tn−1 Tn−2 · · · T1 ) + 1 = 1 + T1 · · · Tn . Then take m < n. Therefore, Tm divides T1 · · · Tn−1 = Tn −1 and thus gcd(Tm , Tn ) = 1. Further, we prove by induction that n 1 1 =1− . Ti Tn+1 − 1 i=1

In fact, k+1 1 1 1 1 1 =1− =1− + =1− . Ti Tk+1 − 1 Tk+1 Tk+1 (Tk+1 − 1) Tk+2 − 1 i=1

Since Tn tends to infinity, we therefore obtain

∞

1 i=1 Ti

= 1.

6.2 Algebraic Combinatorics

99

Problem 2.26. Let α, β > 0 and consider the sequences [α], [2α], . . . , [kα], . . . ; [β], [2β], . . . , [kβ], . . . , where the brackets denote the integer part. Prove that these two sequences taken together enumerate Z+ in an injective manner if and only if α, β ∈ R \ Q and

1 1 + = 1. α β

Solution 2.26. Set AN = {1, 2, . . . , N}; take k maximal such that kα < N + 1 and l maximal such that the following inequalities hold: 1. Therefore, lβ < N + N N+1 N N+1 ≤ l ≤ . ≤ k ≤ and α α β β Since AN is injectively enumerated by the two sequences, we have k + l = N , and hence

N N N +1 N +1 + ≤N ≤ + . α β α β Letting N go to infinity, we obtain α1 + β1 = 1. p If α ∈ Q, then also β ∈ Q. In this case, write α = m n and β = q . It follows that [αnqp] = [βmqn], which contradicts the injectivity assumption. Therefore α ∈ Q and β ∈ Q. Conversely, we have

N +1 N +1 N +1 N +1 N +1 N +1 + > + > + −2 = N −1, N +1 = α β α β α β

whence

N +1 N +1 + = N. α β

In particular, using the notation from above, we have k + l = N . If the two sequences enumerate AN in a surjective manner, then they will also enumerate AN injectively, because k +l = N . It suffices then to prove the surjectivity. Let us assume the contrary. Then there exist u, x, y ∈ Z+ such that xα < u < u + 1 < (x + 1)α and yβ < u < u + 1 < (y + 1)β. Dividing by α, β respectively and adding up the inequalities, we obtain x+y

1 and α1 , . . . , αm ∈ R+ . Then the sets

100

6 Algebra and Combinatorics Solutions

Si = {[nαi ], where n ∈ Z+ } form a complementary system only if m = 2, α1−1 + α2−1 = 1, and α1 ∈ R \ Q. Solution 2.27. If all αj < 1, then the collection Sj contains twice some number. Let 1 ∈ S1 , and thus [α1 ] = 1, α1 = 1. We prove first that S1 = {[nα1 ]; n ∈ Z+ } and T = {[nβ − ]; n ∈ Z+ } form a complementary system, where ' α1 (2(a − c))−1 , if α1 = ac ∈ Q, β= , and = 0, otherwise. α1 − 1 Let = {nα1 , nβ − ; n ∈ Z+ }. It is sufficient to show that for any integer M > 1, we have the formula N = card({x ∈ ; x < M}) = M − 1. Now, nα1 < M is equivalent to nα1 + δ < M, where δ = (2c)−1 if α1 = a/c ∈ Q, while δ = 0, if α ∈ R \ Q. The maximum value for n is therefore M−δ α1 . In a similar way, the number of elements of the form nβ − that are less than M is M+ . This β implies that N = [(M − δ)/α1 ] + [(M + )/β]. Now we have

M −δ M −δ M −δ < −1< , α1 α1 α1

M + M + M + −1< < , β β β and by summing up the two inequalities, we obtain M − 2 < N < M, and therefore N = M − 1, as claimed. Now let k be the smallest integer k ∈ S1 . Then there exists some α2 such that k = [α2 ] = [β − ] ≥ 2. We have nβ − 1 ≤ [nβ − ] < nβ − , and therefore [(n + 1)β − ] − [nβ − ] < (n + 1)β − − nβ + 1 = β + 1 − ≤ k + 2 − , [(n + 1)β − ] − [nβ − ] > (n + 1)β − 1 − nβ + = β − 1 + > k − 1 + 2 . Therefore k ≤ [(n + 1)β − ] − [nβ − ] = [(n + 1)α2 ] − [nα2 ] ≤ k + 1. The difference between two consecutive terms of the sequence [nβ − ] is equal to the difference between consecutive missing terms from the sequence [nα1 ]; that is, k or k + 1 is the same as the difference between two consecutive terms of the sequence [nα2 ]. This implies the fact that the j th term that does not belong to S1 is precisely [j α2 ] = [jβ − ]. This implies that m = 2. If α1 ∈ Q, then α2 = b/d, and for j = d, we obtain [j α2 ] > [jβ − ], which is false.

6.2 Algebraic Combinatorics

101

Comments 40 The fact that the two sequences from the previous problem are complementary is known as the Rayleigh–Beatty theorem, since S. Beatty proposed it as a problem in Amer. Math. Monthly in 1926. Presumably, this was known to Rayleigh, who mentioned it without proof in 1894. In 1927, J.V. Uspensky proved that if the m sequences are complementary, then m ≤ 2; his proof was simplified in several papers by Skolem, Graham, and Fraenkel, who also provided a far-reaching generalization in 1969. • • • • • • •

S. Beatty: Problem 3173, Amer. Math. Monthly 33 (1926), 3, 156. J. Lambek, L. Moser: Inverse and complementary sequences of natural numbers, Amer. Math. Monthly 61 (1954), 454–458. A.S. Fraenkel: Complementary systems of integers, Amer. Math. Monthly 84 (1977), 114–115. A.S. Fraenkel: The bracket function and complementary sets of integers, Canad. J. Math. 21 (1969), 6–27. R.L. Graham: On a theorem of Uspensky, Amer. Math. Monthly 70 (1963), 407– 409. Th. Skolem: On certain distributions of integers in pairs with given differences, Math. Scand. 5 (1957), 57–68. J.V. Uspensky, M.A. Heaslet: Elementary Number Theory, McGraw-Hill, New York, 1939.

Problem 2.28. Let f : Z+ → Z+ be an increasing function and set F (n) = f (n) + n, G(n) = f ∗ (n) + n, where f ∗ (n) = card({x ∈ Z+ ; 0 ≤ f (x) < n}). Then {F (n); n ∈ Z+ } and {G(n); n ∈ Z+ } are complementary sequences. Conversely, any two complementary sequences can be obtained this way using some nondecreasing function f . Solution 2.28. We compute the number N of integers from {F (n), G(n), n ∈ Z+ } that are smaller than M. Suppose that k terms of the form G(n) are smaller than M. Then f ∗ (k) + k < M ≤ f ∗ (k + 1) + k + 1, and so f ∗ (k) < M − k < f ∗ (k + 1) + 1. Since f ∗ (k) = card{x ∈ Z+ ; f (x) < k} < M − k, we derive f (M − k) ≥ k. Further, f ∗ (k + 1) = {x ∈ Z+ ; f (x) < k + 1} ≥ M − k − 1 implies that f (M − k − 1) < k + 1. Therefore F (M − k) = f (M − k) + M − k ≥ M and F (M − k − 1) < M. This means that exactly M − k terms of the form F (n) are smaller than M; therefore N = M. The converse is immediate by defining f (n) = F (n) − n. Comments 41 Since f ∗ is nondecreasing, it follows that it makes sense to define f ∗∗ . The sequences G(n) and H (n), where H (n) = f ∗∗ (n) + n, are complementary, and thus f ∗∗ = f . The result is due to A. Frankel.

102

6 Algebra and Combinatorics Solutions

Problem 2.29. Let M denote the set of bijective functions f : Z+ → Z+ . Prove that there is no bijective function between M and Z. Solution 2.29. We will find an injection ρ : R \ (Q ∪ [0, 2]) → M, as follows. To any α ∈ R \ (Q ∪ [0, 2]) there is associated an irrational β < 2 such that α1 + β1 = 1. Let ρ(α) : Z+ → Z+ be the function defined by ' n α , for even n, ρ(α)(n) = 2n+1 β 2 , for odd n. According to the previous problem, ρ(α) : Z+ → Z+ is a bijection. Moreover, the map ρ is easily seen to be injective. Thus card(M) ≥ card(R \ Q) = card(R) > card(Z), and the result follows. Problem 2.30. Let F ⊂ Z be a finite set of integers satisfying the following properties: 1. For any x ∈ F , there exist y, z ∈ F such that x = y + z. 2. There exists n such that, for any natural number 1 ≤ k ≤ n, and any choice of x1 , . . . , xk ∈ F , their sum x1 + · · · + xk is nonzero. Prove that card(F ) ≥ 2n + 2. Solution 2.30. By hypothesis, 0 ∈ F . Let F+ = F ∩ Z+ , F− = F ∩ Z− , so that F = F+ ∪ F− . Consider the unoriented graph whose vertices are the elements of F+ and whose edges are defined below: x and y are adjacent if there exists z ∈ F such that x = y + z. By the first hypothesis, each vertex is adjacent to at least one edge. This implies that the graph contains a cycle [x1 , . . . , xk ]. Assume that k is minimal with this property. This means that there exist zi ∈ F such that: x1 = x2 + z1 = x3 + z2 + z1 = · · · = x1 + zk + · · · + z1 , which implies that z1 + · · · + zk = 0. The second hypothesis implies that k ≥ n + 1, and thus card(F+ ) ≥ k ≥ n + 1. A similar argument shows that card(F− ) ≥ n + 1, and the claim follows. Problem 2.31. For a finite graph G we denote by Z(G) the minimal number of colors needed to color all its vertices such that adjacent vertices have different colors. This is also called the chromatic number of G. Prove that the inequality Z(G) ≥ holds if G has p vertices and q edges.

p2

p2 − 2q

6.2 Algebraic Combinatorics

103

Solution 2.31. If we fix N = χ (G) and the number p of vertices of the graph G, then we have to show that the number q of edges allowed p2 1 q≤ 1− . 2 N By hypothesis, there exists a partition (according to the colors) of the set of vertices into N classes such that two vertices in the same class are not adjacent. Let n1 , n2 , . . . , nN be the number of vertices in the respective classes. The only possibility to get an edge is to join two vertices lying in different classes, and thus the total number q of edges is at most 1≤i<j ≤N ni nj . Further, one knows that 1≤i≤N ni = p. We now have

ni nj ≤

2 CN

i<j

p2 p2 = N2 2

1 1− N

.

The equality holds if and only if N divides p and ni = p/N , for all i. Comments 42 It can be proved, in the same way, that we have p t − [t] χ (G) ≥ 1− , [t] 1 + [t] where t = p −

2q p .

There exists also an upper bound for an arbitrary graph, namely χ (G) ≤

1 1 + 8q + 1 . 2

One problem that received a great deal of attention in the past was to get bounds for the chromatic number of a graph in terms of its topology. For instance, assume that the graph G is drawn on some surface so that its edges are not crossing each other, in which case the graph is said to be embedded. If the surface is partitioned into curvilinear polygons (called “countries”), then the dual graph of this decomposition is constructed by associating a vertex to each country and joining two vertices if the respective countries have a common frontier. This way one constructs all graphs embedded in that surface. Surfaces in R3 are characterized by only one number, called the genus g, or equivalently, by its Euler–Poincaré characteristic, which is 2 − 2g. The latter is computed elementarily starting from an arbitrary partition into curvilinear polygons. If such a partition consists of V vertices, E edges, and F polygonal countries, then 2 − 2g = V − E + F. Computing a bound for the chromatic number of an arbitrary graph embedded in a surface is equivalent to finding the number of colors needed for coloring any map of the surface so that adjacent countries have different colors. Now, if G is a graph that can be embedded in a surface of genus g, then it was stated without proof by Heawood in 1890 that

104

6 Algebra and Combinatorics Solutions

χ (G) ≤

1 (7 + 1 + 48g) . 2

This has been known as the Heawood conjecture since then, and it was finally proved in 1968 for all surfaces except for the plane by Ringel and Youngs. The case of the plane remained open until 1977, when Appel and Haken presented a computer-assisted proof of that case, thenceforth known as the “four color theorem.” The article of Stahl presents the history of the problem and of the proofs. •

• •

K. Appel, W. Haken: Every Planar Map is Four Colorable, With the collaboration of J. Koch, Contemporary Mathematics, 98, American Mathematical Society, Providence, RI, 1989. G. Ringel: Map Color Theorem, Die Grundlehren der mathematischen Wissenschaften, Band 209, Springer-Verlag, 1974. S. Stahl: The other map coloring theorem, Math. Magazine 58 (1985), 3, 131–145.

Problem 2.32. Let Dk be a collection of subsets of the set {1, . . . , n} with the property that whenever A = B ∈ Dk , then card(A ∩ B) ≤ k, where 0 ≤ k ≤ n − 1. Prove that card(Dk ) ≤ Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . Solution 2.32. Let Pk be the set of those subsets of {1, . . . , n} with at most k elements. Then any maximal (with respect to inclusion) collection Dk as above must contain Pk , since otherwise, Dk ∪ Pk will still satisfy the requirements while being strictly *k = Dk \ Pk . For larger than Dk . Therefore, let us consider a maximal Dk and set D + = {A1 , . . . , As | card(Ai ) = k + 1, and Ai ⊂ A}. If B ∈ D *k *k , let us set D A∈D +∩ B + = ∅, then there exist i, j such that Ai = Bj , where card(Ai ) = k + 1; and A this would imply that A ∩ B ⊃ Ai and thus card(A ∩ B) ≥ k + 1, which is absurd. + for A ∈ D *k , are disjoint. Since each A + has at least one Therefore, the sets A, *k ) ≤ card (Pk+1 \ Pk ) and hence + ⊂ Pk+1 \ Pk , we obtain that card(D element and A *k ) + card(Pk ) = card(Pk+1 ) = Cn0 + Cn1 + Cn2 + · · · + Cnk+1 . We card(Dk ) ≤ card(D have equality for Dk = Pk+1 . Problem 2.33. Prove that

⎧ 0, ⎪ ⎪ ⎪ ⎪ 1, ⎪ ⎪ ⎪ ⎨ n/2,

if if n if 1 (−1)n−k Cnk k p = n(3n+1) , if ⎪ p! 24 ⎪ k=0 ⎪ n2 (n+1) ⎪ ⎪ if ⎪ 48 , ⎪ ⎩ n(15n3 +30n2 +5n+1) , if 1152

0 ≤ p < n, p = n, p = n + 1, p = n + 2, p = n + 3, p = n + 4.

Solution 2.33. Consider the function (ex −1)n , in which we develop first the binomial and then we develop the exponentials in Taylor series:

6.2 Algebraic Combinatorics

(ex − 1)n =

n

(−1)n−k Cnk ekx =

k=0

n

(−1)n−k Cnk

∞ j =0

k=0

n ∞ 1 n−k k j (−1) Cn k . = j! j =0

1 j j k x j!

105

k=0

For the same function we now develop the exponential and then the binomial as follows: n 1 1 (ex − 1)n = x + x 2 + x 3 + · · · 2! 3! n n+1 n(3n + 1) n+2 n2 (n + 1) n+3 x x x + + 2 24 48 n(15n3 + 30n2 + 5n + 1) n+4 + + ··· . x 1152

= xn +

By identifying the first coefficients in the two series, we obtain the formulas from the statement. Comments 43 The problem of computing the sums sn,p = nk=0 (−1)k Cnk k p is a classical one, and the first result (for p = n) is due to Leonhard Euler. D. Andrica gave a method of computation by means of the recurrence relations sn,p+1 = n(sn,p − sn−1,p ),

p ≥ 0.

In particular, one recovers the results from the problem. •

D. Andrica: On a combinatorial sum, Gazeta Mat. 5 (1989), 158.

Problem 2.34. Write lcm(a1 , . . . , an ) in terms of the various gcd(ai , . . . , aj ) for subsets of {a1 , . . . , an }. Solution 2.34. Let Pj = 1≤i1 · · · > σ (n). Let us prove first the following intermediate result. If a < c, b < d, then the inequality |a − b| + |c − d| ≤ |a − d| + |b − c| holds with equality when either a < c < b < d or b < d < a < c. From the symmetry of the previous inequality, we can assume that a < b, d, and using a translation followed by a homothety on the real axis, we can, moreover, assume that a = 0, c = 1. If b < 1, d < 1, the inequality reduces to 2b < 2d. If b < 1, d > 1, then the inequality is 2b < 2. If b > 1, d > 1, we obtain equality. This proves the claim.

6.2 Algebraic Combinatorics

107

Now let σ be a permutation such that S(σ ) is maximal. Suppose that there exists i < j such that σ (i) < σ (j ). Let us define another permutation σ ∗ = σ ◦ (i, j ), where (i, j ) denotes the transposition interchanging i and j . According to our claim, we have S(σ ∗ ) ≥ S(σ ). By modifying iteratively σ by taking the product with the transpositions determined by all pairs (i, j ) as above, we will finally obtain γ : 1 2 ··· n γ : . n n − 1 ··· 1 Moreover, the inequality above shows that S(σ ) ≤ S(γ ) =

n

|n − 2i + 1|.

i=1

Problem 2.37. On the set Sn of permutations of {1, . . . , n} we define an invariant distance function by means of the formula d(σ, τ ) =

n

|σ (i) − τ (i)|.

i=1

What are the values that d could possibly take? Solution 2.37. We have d(ρσ, ρτ ) = d(σ, τ ), for any ρ, σ, τ ∈ Sn . Therefore, it suffices to consider the values of d(1, σ ), where 1 is the identity permutation. If m > 0 is a value of d, we will show below that m − 2 is also a value of d, and hence d takes all even values from 0 to some 2t, where t has to be determined. Notice first that d takes only even values, because d(σ, τ ) ≡

n i=1

(σ (i) − τ (i)) ≡

n i=1

σ (i) −

n

τ (i) ≡ 0 (mod 2).

i=1

Let m = d(1, σ ) > 0. There then exist 1 ≤ r < s ≤ n such that σ (r) > r, σ (s) < s, and σ (i) = i, if r < i < s. Let ρr,s be the cycle (r, r + 1, . . . , s). We claim that d(1, ρr,s σ ) = m − 2. First, by hypothesis, σ (r) > s and σ (s) < r. Looking at the contribution of the elements from r to s to the respective distances, we obtain d(1, σ )−d(1, ρr,s σ ) = (σ (r)−r +s−σ (s))−(r −σ (s)+s−r −1+σ (r)−r −1) = 2. 2 Let us show now that t = n4 (see also the previous problem) and the maximum distance d(1, σ ) is realized θ (i) = n + 1 − i. 2for the permutation 2 We have d(1, θ) = 2 n4 = n2 . Let k < n − 1, such that σ (i) = n + 1 − i, for all i < k, while σ (k) = n+1−k. We claim that there exists some permutation σ˜ ∈ Sn with the property that d(1, σ ) ≤ d(1, σ˜ ) and that satisfies σ˜ (i) = n + 1 − i, i ≤ k. Then, using induction k, we find a sequence of permutations reaching θ such that

108

6 Algebra and Combinatorics Solutions

d(1, σ ) ≤ d(1, σ˜ ) ≤ · · · ≤ d(1, θ). If σ (k) = n + 1 − k, we have σ (k) < n + 1 − k because every number greater than n + 1 − k is the image of some i < k. But this implies that σ (r) = n + 1 − k, for some r > k. Let τ be the transposition (k, r). Set σ˜ = τ σ ; we have d(1, σ˜ ) − d(1, σ ) = |r − kσ | + |m + 1 − 2k| − |k − kσ | + |n + 1 − k − r| = w. Let us suppose that k ≤ σ (k) ≤ r ≤ n + 1 − k (the other cases are similar). Then w = r − σ (k) + n + 1 − 2k − σ (k) + k − n − 1 + k + r = 2(r − kσ ) ≥ 0. Analogously, for all k, r we have d(1, σ˜ ) − d(1, σ ) ≥ 0, which proves the claim. Problem 2.38. The set M = {1, 2, . . . , 2n} is partitioned into k sets M1 , . . . , Mk , where n ≥ k 3 + k. Show that there exist i, j ∈ {1, . . . , k} for which we can find k + 1 distinct even numbers 2r1 , . . . , 2rk+1 ∈ Mi with the property that 2r1 − 1, . . . , 2rk+1 − 1 ∈ Mj . 2 Solution 2.38. There exists a set Ms that contains at least 2n k ≥ 2(k + 1) elements. We have to consider two cases: 2 1. Either Ms contains at least 2(k 2+1) = k 2 + 1 even numbers. Then the set of odd numbers O = {r − 1, where r is even, and r ∈ Ms }

has k 2 + 1 elements. Then there exists some set Ma containing at least k k+1 elements from O. We choose therefore i = s and j = a. Notice that i might be equal to j . 2. Or else Ms contains at least k 2 + 1 odd numbers. The solution is similar to that from above, considering the set of even numbers. 2

Problem 2.39. Let S be the set of odd integers not divisible by 5 and smaller than 30m, where m ∈ Z∗+ . Find the smallest k such that every subset A ⊂ S of k elements contains two distinct integers, one of which divides the other. Solution 2.39. Consider the subset N = {1, 7, 11, 13, 17, 19, 23, 29, . . . , 30m − 1} of elements of S that are not divisible by 3. There are 8m elements in N , which are written in increasing order a1 < a2 < · · · < a8m . Every element of S can be uniquely written as x = ai · 3t , where t ∈ Z+ and ai ∈ N . If k ≥ 8m + 1, then according to the pigeonhole principle, any subset A ⊂ S of cardinality card A = k contains two distinct elements x, y ∈ A for which x = ai 3t and y = ai 3q . In this case, either x divides y, or y divides x. Next, for all i, choose the maximal t (i) with the property that ai 3t (i) < 30m < t ai 3 (i)+1 , and set bi = ai 3t (i) . We have therefore 10m < bi < 30m. The set {b1 , b2 , . . . , b8m } contains 8m elements from S, and also we have the inequalities 0 < bi /bj < 3, for any i, j . Since all numbers bi are odd, we derive that bi /bj ∈ Z. Therefore {b1 , . . . , b8m } does not contain a number and one divisor of it. This shows that the required value of k is 8m + 1.

6.2 Algebraic Combinatorics

Problem 2.40. Prove that · · · ≤ an are integers.

ai −aj 1≤j j (ai − aj ) S= = . i−j (n − 1)!(n − 2)! · · · 1! i>j

We now consider the determinant D of the matrix ⎛ 1 1 ··· 1 ⎜ Ca1 Ca1 · · · Ca1 n 1 2 ⎜ ⎜ .. .. .. ⎝ . . .

⎞ ⎟ ⎟ ⎟. ⎠

Can−1 Can−1 · · · Can−1 n 1 2 k using factorials, It is obvious that D ∈ Z+ . We now write the binomial coefficient Cm and we arrange the factors in the lines. We obtain ⎞ ⎛ 1 1 ··· 1 ⎜ a1 a2 · · · an ⎟ ai − aj 1 i>j ⎟ ⎜ det ⎜ .. = S, D= .. .. ⎟ = ⎠ ⎝ . 1!2! · · · (n − 1)! . . i=1 (i − 1)! a1n−1 a2n−1 · · · ann−1

and the claim follows. Problem 2.41. Is there an infinite set A ⊂ Z+ such that for all x, y ∈ A neither x nor x + y is a perfect power, i.e., a k , for k ≥ 2? More generally, is there an infinite set A ⊂ Z+ such that for any nonempty finite collection xi ∈ A, i ∈ J , the sum i∈J xi is not a perfect power? Solution 2.41. 1. Yes. Let p1 = 2, p2 = 3, . . . , pn , . . . be the set of primes in increasing order. We then set A = {22 3, 22 32 5, . . . , 22 32 52 · · · pn2 pn+1 , . . .}. Obviously, no element of A is a perfect power. Moreover, if x ≤ y ∈ A, then we can write x = 22 · · · pk2 pk+1 and x = 22 · · · pn2 pn+1 . If k < n, then pk+1 divides 2 2 · · · pn2 pn+1 , while pk+1 does not divide x + y, x + y = 22 · · · pk2 pk+1 1 + pk+1 pk+2 and thus it cannot be a perfect power. If k = n, the same argument works, unless k = 0, which was excluded from A. 2. Let us prove that if C ⊂ Z+ is a set of density d(C) = 0, then there exists an infinite set A ⊂ Z+ such that for any nonempty finite collection xi ∈ A, i ∈ J , the sum i∈J xi is not in C. Recall that the density of the subset C ⊂ Z+ is defined as d(C) = lim

n→∞

card(C ∩ {1, 2, . . . , n}) . n

Since d(C) = 0, there exists a natural number a1 ∈ C. Then consider C1 = C ∪ (C − a1 ) ∩ Z+ , where we set X − λ = {x − λ|x ∈ X}. The density d has the following fundamental property that results from the definition:

110

6 Algebra and Combinatorics Solutions

d(A ∪ B) ≤ d(A) + d(B). This implies that d(C1 ) = 0, and therefore there exists a natural number a2 ∈ C1 . Assuming that both the subset Ck of density zero and the sequence a1 , a2 , . . . , ak+1 , with ak+1 ∈ Ck , are defined, we set Ck+1 = Ck

k

(Ck − (ai1 + ai2 + · · · + aip )) ∩ Z+ .

p=1 i1 ≤i2 ≤···≤ip ≤k+1

It follows that d(Ck+1 ) = 0 and thus there exists at least one integer ak+2 > ak+1 such that ak+2 ∈ Ck+1 . The sequence (ak ) forms an infinite subset A with the required property. It suffices now to compute the density of the set C = {a k |a ≥ 2, k ≥ 2}. We have √ √ √ [ n] + [ 3 n] + · · · log n n log2 n card(C ∩ {1, 2, . . . , n}) = ≤ = √2 . n n n n Therefore d(C) = 0. Comments 44 There is an analogous result when products are used instead of sums of elements in A. For instance, the set ( ) an a1 a2 A = a1 , a2 = C2a , C , . . . , C a +a a +a n 1 2 n−1 1 has the property that no product of its elements is a perfect power. Is it true that d(A) = 0? Problem 2.42. Let .. A A2 3

f (n) = max A1

Ak

. ,

where n = A1 + · · · + Ak . Thus, f (1) = 1, f (2) = 2, f (3) = 3, f (4) = 4, f (5) = 9, f (6) = 27, f (7) = 512, etc. Determine f (n). A ·· k

Solution 2.42. It is clear that if f (n) = A·1

, then Ai > 1. Also, f is an increasing A ·· k

function, and thus f (n + 1) ≥ f (n) + 1. If A1 = k, then A·2 implies that whenever n ≥ 4, we have f (n) =

≤ f (n − k), and this

max k f (n−k) .

2≤k≤n−2

It is easy now to see that f (n + 1) ≥ 2f (n). In fact, let f (n) = k f (n−k) , for some k ≥ 2; then f (n + 1) ≥ k f (n−k+1) ≥ k f (n−k)+1 ≥ 2f (n). log(a+1) The next step is to compare a f (b+1) with (a + 1)f (b) , that is, f f(b+1) (b) with log(a) . If a ≥ 2, then a + 1 < a 2 and hence log(a + 1)/ log a < 2. Thus, if b ≥ 4, then

6.2 Algebraic Combinatorics

111

f (b + 1)/f (b) ≥ 2, which yields a f (b+1) > (a + 1)f (b) , as soon as a ≥ 2, b ≥ 3. For small values of the arguments, we have log 4 f (4) f (3) log 3 < < < . log 3 f (3) f (2) log 2 These show that k f (n−k) has a maximum when k = 2 if n > 6. Therefore, the final answer is ⎧ 2 23 ⎪ ⎨ ··· 2 , for odd n > 3, f (n) = 3 23 ⎪ · ⎩ ·· 2 , for even n > 4. Problem 2.43. Consider a set M with m elements and A1 , . . . , An distinct subsets of M such that card(Ai ∩ Aj ) = r ≥ 1 for all 1 ≤ i = j ≤ n. Prove that n ≤ m. Solution 2.43. Observe first that if A is an m×n matrix and m > n, then det A·A = 0, where A denotes the transpose matrix. In fact, let us border the matrix A by a null matrix of size (m − n) × n, in order to get an matrix A . It is obvious that m × m A · A = A · A , and thus det A · A = det A · A = 0. Let D be the n × n matrix whose entries are ' r ≥ 1, if i = j, Dij = dii ≥ r, otherwise. Assume that there exists at most one i such that dii = r. We claim then that det(D) = 0. This follows by subtracting the last line from the first n − 1 and an inductive argument. Consider now the m × n matrix A whose entries are ' 1, if i ∈ Aj , Aij = 0, otherwise. It is now immediate that D = A · A is given by Dij =

n

aik akj = card(Ai ∩ Aj ).

j =1

By hypothesis, Dij = r when i = j , and since the Ai are distinct, there exists at most one i such that Dii = r. The claim from above tells us that det(D) = det A · A = 0, while the first observation implies that m ≤ n. Problem 2.44. Set π(n) for the number of prime numbers less than or equal to n. Prove that there are at most π(n) numbers 1 < a1 < · · · < ak ≤ n with gcd(ai , aj ) = 1.

112

6 Algebra and Combinatorics Solutions

Solution 2.44. If X is a set, we denote by P(X) the set of nonempty subsets of X. Define the map h : {1, . . . , n} → P({p1 , . . . , pπ(n) }) that associates to the integer x the set of prime divisors of x. One then has card(h(ai )) ≥ 1 and further h(ai )∩h(aj ) = ∅, from our assumption. Since the maximal number of disjoint subsets of {p1 , . . . , pπ(n) } is bounded by π(n), the claim follows. Problem 2.45. Prove that for every k, there exists n such that the nth term of the Fibonacci sequence Fn is divisible by k. Recall that Fn is determined by the recurrence Fn+2 = Fn + Fn+1 , for n ≥ 0, where the first terms are F0 = 0, F1 = 1. Solution 2.45. Using the pigeonhole principle, there exists an infinite sequence ni such that Fn1 ≡ Fn2 ≡ · · · ≡ Fnm ≡ · · · (mod k). Moreover, there exists an infinite subsequence nis such that Fni1 +1 ≡ Fni2 +1 ≡ · · · ≡ Fnim +1 ≡ · · · (mod k) by the same argument. Thus one knows that Fnis ≡ Fnit (mod k) and Fnis +1 ≡ Fnit +1 (mod k). The recurrence relation of the Fibonacci numbers shows that Fnis +m ≡ Fnit +m (mod k), for any m ∈ Z. Since F0 = 0, we find that there exists n (actually infinitely many such integers) such that k divides Fn . Problem 2.46. Consider a set of consecutive integers C + 1, C + 2, . . . , C + n, where C > nn−1 . Show that there exist distinct prime numbers p1 , p2 , . . . , pn such that C + j is divisible by pj . Solution 2.46. Let k ≤ h and consider the factorization of C + k into prime factors. If C + k has at least n distinct prime divisors, then we can find one prime divisor that is not yet associated with the other n − 1 numbers C + i. Assume then that C + k has at most j ≤ n − 1 prime factors, which are p1k , . . . , pj k for 1 ≤ j ≤ n − 1. We write down the factorization as a

a1k · · · pj jkk = C + k > C > nn−1 . p1k a

The pigeonhole principle implies that there exists some prime power qk = pj jkk such that qk > n. We then associate the prime number pj k to C + k. Let us show that this procedure yields distinct prime numbers. Suppose that there exist j and k such that the associated numbers are the same. We know that qk divides C + k and qj divides C + j , and they are both prime powers of the same prime. Since qk > n and qj > n, we obtain gcd(qk , qj ) > n. Moreover, gcd(qk , qj ) divides both C + k and C + j and hence their difference |j − k| < n, which is a contradiction. This proves the claim.

6.2 Algebraic Combinatorics

113

Problem 2.47. Let p be a prime number and f (p) the smallest integer for which there exists a partition of the set {2, 3, . . . , p} into f (p) classes such that whenever a1 , . . . , ak belong to the same class of the partition, the equation k

xi ai = p

i=1

does not have solutions in nonnegative integers. Estimate f (p). Solution 2.47. 1. Suppose that a1 , a2 are in the same class and gcd(a1 , a2 ) = 1. One knows then that any natural number greater than or equal to a1 a2 can be written as √ x1 a1 + x2 a2 , where xi ∈ Z+ . Therefore, the prime numbers smaller than p should to different classes of our partition, and hence √ 1 2 p f (p) > π p 2 ∼ , log p where π(n) denotes the number of primes smaller than n. + , let us consider the sets of consecutive elements At = ( 2. If t ∈ Z p ) p . We claim that the equation ki=1 ai xi = p has no integer t+1 + 1, . . . , t solutions if ai ∈ At for 1 ≤ i ≤ k. In fact, if we had a solution xi , then we would have k k k p p xi < ai xi < xi , t +1 t i=1

k

i=1

i=1

and so t < i=1 xi < t + 1, which is false, since xi are integers. Let us consider the classes A1 , A2 , . . . , AL−1 , where L is an integer to be fixed later. Consider now, for every prime q < p/L, the classes Bq = {q, 2q, . . . , αq, . . .}. It is immediate that the associated linear equation has no solutions if the coefficients belong to some Bq , since q does not divide p. Further, A1 , . . . , AL−1 , B2 , B3 , . . . , B p L form a partition of {2, 3, . . . , p}. The total number of classes of this partition is then L − 1 + π(p/L). " " Setting L =

2p log p ,

we obtain f (p)

0, one can use finitely many foldings of the paper along its sides in either 2 equal parts or 3 equal parts to obtain a rectangle whose sides are in ratio r for some r satisfying 1 − ≤ r ≤ 1 + . Solution 3.19. Observe that log3 2 ∈ Q and thus the residues modulo 1 of the elements n log3 2, with n ∈ Z+ , form a dense subset of (0, 1). This implies that the set of numbers of the form 2a 3b , with a, b ∈ Z, are dense in R. Problem 3.20. Show that there exist at most√three points on the unit disk with the distance between any two being greater than 2.

7.2 Combinatorial Geometry

143

Solution 3.20. Consider the points Ai on the unit√disk of radius 1 and center O. Then |OAi | ≤ 1, and if we assume that |Ai Aj | > 2, then 2 < |OAi |2 + |OAj |2 − π 2|OAi | · |OAj | cos A i OAj . Thus, cos A i OAj < 0 and hence A i OAj > 2 , which implies that n ≤ 3. Comments 56 H.S.M. Coxeter proposed this problem in 1933. The same argument the unit disk, then there exists a pair of points shows that if we have k points Ai within such that |Ai Aj | ≤ max 1, 2 sin πk , and this is sharp for 2 ≤ k ≤ 7. The result was generalized to higher dimensions by Davenport, Hajós, and, independently, Rankin, as follows: There are at most n + 1 points√in the unit ball of Rn such that the distance between any two points is greater than 2. Here is a proof sketch. Let Ai denote the m points satisfying the hypothesis, so that OAi they are different from the origin O. Then we have m ≥ n + 2 unit vectors |OA i| n in R with the property that vi , vj < 0. We claim next that there exist pairwise orthogonal subspaces W1 , W2 , . . . , Wm−n ⊂ Rn that span Rn such that each space Wj contains 1 + dim Wj vectors among the unit vectors above, v1 , . . . , vm , that span Wj . Since m − n ≥ 2, there exist two vectors among the vi that are orthogonal to each other, which contradicts our assumptions. The claim follows by induction on the dimension n. If n = 2, it is immediate. Now, if vi = −vn , for some i < n, then all other vj should be orthogonal to vn (and to vi ). Otherwise, we can write for all i < n, vi = ui + xi vn , for some scalars xi < 0 and some nonzero vector ui orthogonal to vn . This implies that ui , uj < 0, when i = j . We can apply therefore the induction hypothesis for the subspace orthogonal to vn and the set of vectors vi , i < n, and obtain the subspaces Wj . We define then Wi = Wi , for i ≥ 2, and let W1 be the span of W1 and vn . This proves the claim. n The claim shows that if we have √ m ≥ n + 2 points on the unit sphere in R whose pairwise distances are√at least 2, then m ≤ 2n, and moreover, there exist two points at distance precisely 2.

• •

H.S.M. Coxeter: Amer. Math. Monthly 40 (1933), 192–193. R.A. Rankin: The closest packing of spherical caps in n dimensions, Proc. Glasgow Math.Assoc. 2 (1955), 139–144.

Problem 3.21. A convex polygon with 2n sides has at least n diagonals not parallel to any of its sides. The equality is attained for the regular polygon. Solution 3.21. Let l be a side of the polygon. Then we have at most n − 2 diagonals parallel to l. In fact, let d1 , d2 , . . . , dp be these diagonals. Then the pair of segments (d1 , l) determines a quadrilateral (actually a trapezoid) having three common sides with the polygon. Moreover, all the pairs (d2 , d1 ), . . . , (dp−1 , dp ) determine quadrilaterals, each one containing two sides of the initial polygon. This implies that p ≤ n − 2. The total number of diagonals that are parallel to at least one side is at most 2n(n − 2). But there are 2n(2n−3) diagonals, and hence n(2n − 3) − 2n(n − 2) = n 2 diagonals with the desired property.

144

7 Geometry Solutions

Problem 3.22. Let d be the sum of the lengths of the diagonals of a convex polygon P1 · · · Pn and p its perimeter. Prove that for n ≥ 4, we have n n + 1 d n−3 · · · > αn−2 > αn−1 = p. Let us show first that p < α2 . Let Ai Ai+2 ∩ Ai+1 Ai+3 = {Oi+1 }. Observe that |Ai+1 Oi+1 | ≤ |Ai+1 Oi+2 |, since Ai+1 Ai+2 Oi+1 ≤ Ai+1 Ai+2 Oi+2 . The triangle inequality in Ai Oi Ai+1 further yields |Ai+1 Oi+1 | + |Oi+1 Ai+2 | > |Ai+1 Ai+2 |.

Summing up these inequalities, and using the remark, it follows that p < α2 , as claimed. Let now k < n2 − 1, and set Ai Ai+k ∩ Ai+1 Ai+k+1 = {O}. The triangle inequality again shows that |Ai Ai+k | + |Ai+1 Ai+k+1 | = |Ai O| + |OAi+k | + |Ai+1 O| + |OAi+k+1 | > |Ai+1 Ai+k | + |Ai Ai+k+1 |. By summing up over i, we obtain 2αk > αk+1 + αk−1 ; thus αk+1 − αk < αk − αk−1 < · · · < α2 − α1 , and hence the claim. Moreover, if n = 2m, then 2αm > αm+1 + αm−1 = 2αm+1 , and so αm > αm−1 . Thus, we have proved that αk+1 > αk for all k. By summing up all these inequalities over k, we find that 2d > (n − 3)p.

7.2 Combinatorial Geometry

145

Further, consider the triangle inequality |Ai Aj | < |Ai Ai+1 | + · · · + |Aj −1 Aj | and sum up over all i and j ; this yields

n n+1 2d < − 2 p. 2 2 Problem 3.23. Find the convex polygons with the property that the function D(p), which is the sum of the distances from an interior point p to the sides of the polygon, does not depend on p. Solution 3.23. Let K be our polygon, p an interior point, a1 , . . . , an the distances from p to the sides of K, and let u1 , . . . , un be unit vectors perpendicular to the sides. Choose another point, say q, in the interior of K. Denote by v the vector pq. The distance from q to the side with label i is therefore di = ai + ui , v, where , is the scalar product. It follows that D(q) = D(p) +

- n

. ui , v .

i=1

Therefore, a necessary and sufficient condition for D(p) to be constant is that n

ui = 0.

i=1

Observe that the solution can be extended to convex polyhedra in Rm . Problem 3.24. Prove that a sphere of diameter 1 cannot be covered by n strips of width li if ni=1 li < 1. Prove that a circle of diameter 1 cannot be covered by n strips of width li if ni=1 li < 1. Solution 3.24. 1. The lateral area of the body obtained by cutting off the sphere with two planes at distance h is 2πRh. Let us assume that we have a covering by means of the strips Bi . Thenthe sum of areas cut open by the strips Bi is at least the area of the sphere, hence π ni=1 li ≥ π , contradicting our assumptions. 2. If we had such a covering, let Bi∗ be the spatial strip that cuts the planar strip Bi , orthogonal to the plane of the circle. Then the union of the strips Bi∗ would cover the cylinder over the disk, in particular the sphere of diameter 1. This would contradict the first part. Problem 3.25. Consider n points lying on the unit sphere. Prove that the sum of the squares of the lengths of all segments determined by the n points is less than n2 . Solution 3.25. Let O be the center of the sphere and let Xi denote the given points. −−→ We denote by vj the vector OXj , and by , the Euclidean inner product. We have then |Xi Xj |2 = vj − vi , vj − vi .

146

7 Geometry Solutions

Set S = 2S =

i,j

n

|Xi Xj |2 . We have then

vi − v1 , vi − v1 +

i=1

n i=1

= 2n(|v1 |2 + · · · + |vn |2 ) − 2

vi − v2 , vi − v2 + · · · +

n

vi − vn , vi − vn

i=1

vj , vi

i,j

= 2n

n

|vi |2 − 2v, v = 2n2 − 2|v|2 ≤ 2n2 ,

i=1

where v =

n

i=1 vi .

−−→ −−→ Problem 3.26. The sum of the vectors OA1 , . . . , OAn is zero, and the sum of their lengths is d. Prove that the perimeter of the polygon A1 . . . An is greater than 4d/n. −−→ −→ Solution 3.26. If A is an arbitrary point in the plane, then nOA = ni=1 (OAi + n −−→ −−→ Ai A) = i=1 Ai A, which implies that n|OA| ≤ |Ai A|. Let us put A = Ai for i = 1, 2, . . . , n and sum up the inequalities so obtained: d = |OA1 | + · · · + |OAn | ≤

2 |Ai Aj |. n i,j

Using the triangle inequality, |Ai Aj | < |Ai Ai+1 | + · · · + |Aj −1 Aj |, we infer that d≤

2 n−1 n2 − 1 p + 2p + · · · + p = p. n 2 4n

Therefore p ≥ 4d n2n−1 >

4d n .

Problem 3.27. Find the largest numbers ak , for 1 ≤ k ≤ 7, with the property that for any point P lying in the unit cube with vertices A1 . . . A8 , at least k among the distances |P Aj | to the vertices are greater or equal than ak . Solution 3.27. At least eight distances are greater than or equal to 0, and this is sharp by taking P = Aj , and hence a8 = 0. By taking P to be the midpoint of an edge, we get a7 ≤ 21 . Assume that a7 < 21 . Then there exists some P for which at least two distances |P Aj | are strictly smaller than 21 , which is obviously false, since vertices are distance 1 apart. Thus a7 = 21 . √

Consider P as the center of some face of the cube. We obtain then a6 ≤ a5 ≤ 22 . If equality does not hold above, then there exists P for which three distances are √ 2 smaller than 2 . However, given three vertices of the cube, there are at least two of √ them at distance greater than or equal to the diagonal of a face, namely 2. Thus our √ assumption fails and so a6 = a5 = 22 .

7.2 Combinatorial Geometry

Now let P be the center of the cube. We derive that a4 ≤ a3 ≤ a2 ≤ a1 ≤ √

Assume that a4 < √ 3 2 .

3 2 .

147 √

3 2 .

Then there exist P and at least five distances strictly smaller

However, for any choice of five vertices of a cube, one finds a pair of than √ opposite vertices that are distance 3 apart. This contradicts our assumptions and √ thus a4 = a3 = a2 = a1 = 23 . Problem 3.28. The line determined by two points is said to be admissible if its slope is equal to 0, 1, −1, or ∞. What is the maximum number of admissible lines determined by n points in the plane? Solution 3.28. We join two points by an edge if the line determined by them is admissible. We obtain a graph with n vertices. By hypothesis, each vertex has degree at most 4. Since the sum of degrees over all vertices is twice the number of edges, we find that there are at most 2n edges, thus at most 2n admissible lines. Moreover, there are only four directions in the plane, and the same computation shows that there are at most 21 n edges associated to every direction. This implies that for odd n, the maximum number of admissible lines is 2n − 2. It is easy to construct examples in which we have precisely 2n admissible lines if n is even and n ≥ 12. For odd n, the associated graph is obtained by adding an isolated vertex to a maximal graph of order n − 1, and this works for any n ≥ 3. Finally, if f (n) denotes the maximum number of admissible lines, then explicit inspection shows that f (2) = 1, f (3) = 3, and f (4) = 6, and so the upper bound 2n is not attained. It can be shown that f (6) = 11 and f (10) = 19. Therefore, for all n but those from {2, 3, 4, 6, 10}, we have f (n) = 4 n2 . Comments 57 More generally, assume that the admissible directions are k in number. The same proof shows that there therefore exist at most kn 2 admissible lines determined by n points. Moreover, the minimum number of directions (not necessarily admissible) defined by n points is at least n − 1. See also the article: •

P.R. Scott: On sets of direction determined by n points, Amer. Math. Monthly 77 (1970), 502–505.

Problem 3.29. If A = {z1 , . . . , zn } ⊂ C, then there exists a subset B ⊂ A such that n ≥ π −1 z |zi |. z∈B

i=1

Solution 3.29. We reorder the numbers z1 , . . . , zn , −(z1 + · · · + zn ) in a sequence w1 , . . . , wn+1 such that the sequence arg w1 , . . . , arg wn+1 is monotonic. In the complex plane, the points w1 , w1 + w2 , . . . , w1 + · · · + wn+1 , are the vertices of a convex polygon P . Set d(P ) for the diameter of P . Since the diameter of a polygon is the distance between two vertices, there exists B ⊂ A such that d(P ) = zi . zi ∈B

148

7 Geometry Solutions

Let p(P ) be the perimeter of P . We have then p(P ) =

|wi | =

n

n |zi | + zi .

i

i

The main isoperimetric inequality states that for any convex planar polygon Q we have the inequality π d(Q) ≥ p(Q). Applying this to the polygon P settles the claim. Comments 58 Using a refined isoperimetric inequality for polygons with a given number of vertices, one finds that there exists B ⊂ {1, . . . , n} such that n sin(kπ/(2k + 1)) ≥ z |zi |. i (2k + 1) sin(π/2k + 1) i∈B

i=1

Problem 3.30. Let A1 , . . . , An be the vertices of a regular n-gon inscribed in a circle of center O. Let B be a point on the arc of circle A1 An and set θ = A n OB. If we set ak = |BAk |, then find the sum n (−1)k ak k=1

in terms of θ . Solution 3.30. We have ak = 2r sin n

(−1)k ak cos

k=1

kπ n

−

θ 2

. Then

θ θ (2k + 1)π π (2k − 1)π = (−1)k r sin − + sin − 2n 2n 2 2n 2 π θ = −1 + (−1)n+1 r sin − , 2n 2

which leads us to n k=1

' (−1) ak = k

π π −1 cos 2n , for even n, 2r sin θ2 − 2n 0, for odd n.

Problem 3.31. Consider n distinct complex numbers zi ∈ C such that min |zi − zj | ≥ max |zi |. i=j

i≤n

What is the greatest possible value of n? Solution 3.31. Let |zn | = maxi≤n |zi |; then the zi belong to the disk of radius R = |zn | centered at the origin. Moreover, the distance between two points of complex coordinates zi and zj is given by |zi − zj |, which by hypothesis is greater than R. In particular, we have n points whose pairwise distances are at least R within the circle of radius R. Let O be the origin and Zi the points. We order the points Zi in

7.2 Combinatorial Geometry

149

increasing order with respect to their arguments. If the origin is among the Zi , then we choose it to be the first point. Further, there remain at least n − 1 points distinct from the origin. One knows that n

Zi OZi+1 = 2π,

i=2 2π OZi+1 ≤ n−1 . In particular, if n ≥ 8, and thus there exists at least one angle Zi " then |Zi Zi+1 | ≤ R 2 − 2 cos Zi OZi+1 < R, contradicting our assumptions. Thus n ≤ 7 and the maximal value n = 7 is reached by the configuration of six vertices of a regular hexagon together with the origin.

Problem 3.32. The interior of a triangle can be tiled by n ≥ 9 pentagonal convex surfaces. What is the minimal value of n such that a triangle can be tiled by n hexagonal strictly convex surfaces? Solution 3.32. There is an obvious recurrence, showing how to go from k to k + 3.

We still have to find the appropriate tiling for n ∈ {9, 10, 11}. If n = 9, then we have the following tiling of the triangle:

A slight deformation of the left corner pentagon shows that a concave quadrilateral can also be tiled with nine pentagons. Moreover, a tiling by n pentagons of a concave quadrilateral can be adjusted as below to a tiling of a triangle with n + 1 pentagons. This method furnishes a tiling with any number n ≥ 9 of pentagons.

150

7 Geometry Solutions

2. We will show that there are no hexagonal tilings of triangles with convex tiles. Assume the contrary. We call a tiling good if no hexagon tile has sides in common with (at least) two sides of the triangle. Given an arbitrary tiling, we can construct a good tiling of a triangle as follows. First, the triangle can be assumed to be a right triangle by means of the following transformation. Consider the cone in R3 over the initial tiled triangle with a cone vertex far away, and cut it by a convenient plane so that one of the angles of the section become right angled. The trace of the coned tiling in the planar section is still a hexagonal tiling with convex polygons. Further, we double the right triangle along one of its sides and obtain a tiling that has no hexagons with common edges with both of the sides opposite to the doubled side. Applying again the same procedure yields a good tiling. The main feature of a good tiling by convex polygons is that any hexagon having a common side with the triangle (called a boundary hexagon) has precisely one side in common, excepting the case in which the hexagon might have a common angle with the triangle (called an exceptional boundary hexagon). In the last case, we have a vertex of the hexagon and its incident sides that are common with the vertex of a triangle and its incident sides. Thus there are λ ∈ {0, 1, 2, 3} exceptional boundary hexagons. Let a∂ be the total number of boundary hexagons and ain the hexagons that are interior, i.e., they have no sides in common with the triangle.

Construct the dual graph associated to the tiling. This is a planar graph whose vertices have degree 6. The edges going outside of the triangle will be called exterior edges of the graph, and the remaining ones interior edges. The number of vertices of this graph is a = ain + a∂ . Counting the sum of degrees of all vertices, we obtain 6a on one side, and 2ein + eo , where ein (respectively eo ) is the number of interior (respectively exterior) edges. Moreover, each exterior edge comes from a unique boundary hexagon, and two different edges come from different hexagons, excepting the exceptional hexagons, which have two outgoing edges. Thus eo = a∂ + λ. This implies that ein = 3a − 21 (a∂ + λ). If we erase the exterior edges of our graph, then we obtain a planar polygon partitioned into polygons (duals of the hexagons). The total number of vertices is a, the number of edges is ein and the number of faces f can be estimated from above as follows. We have ain vertices in the interior of a polygon with a∂ sides. Each polygon has at least three sides (because our initial hexagons were supposed to be

7.2 Combinatorial Geometry

151

strictly convex). Consider the sum of angles of all polygons involved. On the one hand this yields 2πain + π(aδ − 2), and on the other hand this is at least πf . Thus f ≤ 2ain + a∂ − 2. The Euler–Poincaré theorem requires that a∂ 1 λ 1 = a − ein + f ≤ a − 3a − (a∂ + λ) + 2ain + a∂ − 2 = − 2 − . 2 2 2 But λ ≤ 3 and a∂ ≥ 0 lead us to a contradiction. Comments 59 The same proof shows that there does not exist a tiling of the quadrilateral by strictly convex hexagons. Problem 3.33. We say that a transformation of the plane is a congruence if it preserves the length of segments. Two subsets are congruent if there exists a congruence sending one subset onto the other. Show that the unit disk cannot be partitioned into two congruent subsets. Solution 3.33. Let us assume that the disk D can be partitioned as D = A ∪ B, where A and B are congruent. Let p denote the center of D, and assume that p ∈ A. Let F : A → B be a congruence.

Let rs be the diameter that is perpendicular to the line pF (p). For any point x ∈ D, we have |px| ≤ 1. Thus |F (p)F (x)| ≤ 1, for any x ∈ A, and thus |F (p)y| ≤ 1, for any y ∈ B. Now |F (p)r| > 1 and |F (p)s| > 1, and thus r, s ∈ A. Further, |F (r)F (s)| = |rs| = 2 and thus F (r)F (s) is a diameter of the disk. Since |F (r)F (p)| = |F (s)F (p)| = 1, we derive that F (p) is the midpoint of the diameter F (r)F (s), and thus it coincides with the center p. This contradiction proves the claim. Comments 60 Actually, the same argument shows that the disk cannot be partitioned into finitely many congruent sets. The problem is due to B.L. Van der Waerden. More generally, D. Puppe showed that the same holds for any convex compact set in the plane. See also: •

H. Hadwiger and H. Debrunner: Combinatorial Geometry in the Plane, translated by V. Klee. With a new chapter and other additional material supplied by the translator, New York, 1964.

Problem 3.34. Prove that the unit disk cannot be partitioned into two subsets of diameter strictly smaller than 1, where the diameter of a set is the supremum distance between two of its points.

152

7 Geometry Solutions

Solution 3.34. Observe first that the diameter of a planar set A coincides with the diameter of its closure A in R2 . Assume that the disk is partitioned into two sets A and B of smaller diameter. The points of the boundary circle cannot belong to only one of the partition sets, since opposite points are at distance 1. One the other hand, the unit circle cannot be written as the union of two disjoint nontrivial closed subsets (i.e., closed intervals) because it is connected. This implies that there exists some point M of the unit circle that belongs to both A and B. Consider now the opposite point M on the circle. If M ∈ A, then A has diameter 1; otherwise, M ∈ B and so B has diameter 1, which gives us a contradiction. Problem 3.35. A continuous planar curve L has extremities A and B at distance |AB| = 1. Show that, for any natural number n there exists a chord determined by two points C, D ∈ L that is parallel to AB and whose length |CD| equals n1 . Solution 3.35. Assume that L does not admit chords CD parallel to AB either of length a or of length b. We will show that L does not admit chords CD parallel to AB of length a + b. Let Tx be the planar translation along the line AB of the number of x units. We have, by hypothesis, Ta L ∩ L = ∅ and Tb L ∩ L = ∅. This implies that Ta+b L∩Ta L = ∅. On the other hand, let d(U, V ) denote the distance between the sets U and V in the plane, given by inf P ∈U,Q∈V |P Q|. Then obviously d(Ta+b L, L) ≤ d(Ta L, L) + d(Tb+a L, Ta L). This implies that d(Ta+b L, L) = 0 if d(Ta L, L) = d(Tb+a L, Ta L) = 0. Moreover, L and Ta+b (L) are closed subsets, and thus the distance is attained, i.e., there exist P ∈ L and Q ∈ Ta+b (L) such that |P Q| = d(L, Ta+b (L)). In particular, P = Q and thus Ta+b L ∩ L = ∅, proving our claim. Suppose now that there is no parallel chord of length a = b = n1 . Then we derive that there is no such chord of length n2 . Take a = n1 and b = n2 . The previous claim shows that there is no chord of length n3 . An easy induction shows that there is no chord of length nk for k ≤ n. This contradicts the fact that AB has length 1. Comments 61 The result is due to H. Hopf, who characterized all the possible lengths of chords arising as above. •

H. Hopf: Über die Sehnen ebener Kontinuen und die Schleifen geschlossener Wege, Comment. Math. Helv. 9 (1937), 303–319.

Problem 3.36. The diameter of a set is the supremum of the distance between two of its points. Prove that any planar set of unit diameter can be partitioned into three √ 3 parts of diameter no more than 2 . Solution 3.36. The main ingredient is to show that a planar set F of unit diameter is contained within a regular hexagon whose opposite sides are at distance 1. Given a vector v = v1 , let us consider the thinest strip Sv parallel to v that contains F . Then the strip Sv is bounded by two parallel lines, each line containing

7.2 Combinatorial Geometry

153

at least one point of F , by the minimality of the strip. In particular, the width of the strip is bounded by the distance between two points of F , and so by the diameter of F . Now let v2 be the image of v1 by a rotation of angle π3 , and let v3 be the image of the rotation of angle 2π 3 . The intersection of the corresponding strips Hv = Sv1 ∩Sv2 ∩Sv3 is a semiregular (i.e., all opposite edges are parallel) hexagon having all angles 2π 3 . However, in general, this hexagon is not regular.

We choose two opposite vertices, say A and A , of the parallelogram Sv1 ∩ Sv2 . For instance, we choose A such that the vectors spanning the parallelogram issued from it are positive multiples of both v1 and v2 . We denote by x the distance from A to the strip Sv3 and by x the distance from A to the strip Sv3 . We observe that the hexagon Hv is regular if x = x . Moreover, let us set f (v) = x − x . Obviously, the function that associates to the vector v the quantity x − x is continuous. Let us compute f (−v), by rotating v through π. This amounts to interchanging A with A , and thus f (−v) = −f (v). A continuous function taking both positive and negative values must have a zero. Thus there exists some position of v for which f (v) = 0 and so Hv is regular. Further, the regular hexagon of width 1 can be partitioned into three pieces having √ diameter 23 , as in the figure below:

This proves the claim. Comments 62 K. Borsuk proved in 1933 that a planar set of diameter 1 can be partitioned into subsets of diameter d < 1. The main ingredient above is a lemma due to J. Pál, from 1920. B. Grünbaum proved (improving on ideas of D. Gale) that a set of unit diameter in R3 is contained in a regular octahedron of width 1, with three of its vertices (one vertex on each diagonal) cut off by planes orthogonal to the diagonals,

154

7 Geometry Solutions

at distance 21 from the center. Using a suitable dissection of this polyhedron, one finds that the set F can be partitioned into four pieces having diameter not bigger than / √ 6129030 − 937419 3 ≈ 0.9887. √ 1518 2 Gale’s conjecture from 1953, " that one can always get a partition into four pieces of diameter no bigger than • •

√ 3+ 3 6 ,

is still open.

D. Gale: On inscribing n-dimensional sets in a regular n-simplex, Proc. Amer. Math. Soc. 4 (1953), 222–225. B. Grünbaum: A simple proof of Borsuk’s conjecture in three dimensions, Math. Proc. Cambridge Philos. Soc. 53 (1957), 776–778.

Problem 3.37. 1. Prove that a finite set of n points in R3 of unit diameter, can be 2 covered by a cube of side length 1 − 3n(n−1) . 2. Prove that any planar set of n points having unit diameter can be partitioned into √ 3 2π three parts of diameter less than 2 cos 3n(n−1) . Solution 3.37. 1. A direction will mean below a line without specified orientation. The angle between two directions is the among the four angles they define, and

smallest thus it is always within the range 0, π2 . Consider the set of n points {x1 , x2 , . . . , xn }, which determine at most N = n(n−1) distinct directions xi xj , which we denote 2 by l1 , l2 , . . . , lN . We want to prove first that there exist three mutually orthogonal directions y1 , y2 , y3 , so that 1 ∠(yi , ls ) ≥ arccos 1 − . min 1≤i≤3,1≤s≤N 3N Consider an orthogonal frame in R3 given by three mutually orthogonal vectors e1 , e2 , e3 , and choose the rotation R1 and R2 of the space with the property that R1 e1 = e2 , R1 e3 = e3 , R2 e1 = e3 , R2 e2 = e2 . Using these rotations, we have min

1≤i≤3,1≤s≤N

∠(yi , ls ) = min min(∠(y1 , ls ), ∠(y1 , R1 ls ), ∠(y1 , R2 ls )). 1≤s≤N

Consider the set of 3N directions D = {lj , R1 lj , R2 lj , 1 ≤ j ≤ N }. We will prove the following estimate from below: 1 = θ, max min ∠(y, l) ≥ arccos 1 − y l∈D 3N for an arbitrary set D of 3N directions. Assume the contrary. The set of directions y making an angle smaller than θ with a given direction l form a symmetric cone Cl,θ of angle 2θ with axis l. If there is no direction y satisfying the inequality above, then the union of all cones Cl,θ , over l ∈ D will cover the set of all directions in space. Each symmetric cone Cl,θ intersects the unit sphere along a symmetric pair of

7.2 Combinatorial Geometry

155

spherical caps of radius θ. In particular, these spherical caps cover the surface of the whole sphere, since a point not covered yields a direction not belonging to the union. Furthermore, this implies that the total area of the spherical caps is (strictly) greater than the area of the sphere. Now, the area of each spherical cap is 2π(1 − cos θ), and thus the total area covered by the cones is 12πN (1 − cos θ) = 4π . But 4π is the total area of the sphere, which contradicts the claim. Observe that the inequality must be strict, since N ≥ 1 and any covering of the sphere by nontrivial spherical caps will have some overlaps. Consider now three mutually orthogonal directions yj satisfying the claim above. Let further ξj ⊂ yj be the image of the given set of points under orthogonal projection onto yj . The distance between the images of xi and xk on yj is |xi xk | cos ∠(yj , xi xk ) ≤ sin θ . Thus each subset ξj is contained in some interval of length less than cos θ , and thus the set of points is contained in the strip of width sin θ, orthogonal to yj . The intersection of the three orthogonal strips associated to the direction yj is a cube of side length cos θ , and the claim follows. 2. The second part follows using the same idea. This time, we consider three directions y1 , y2 , y3 in the plane such that y2 is obtained by a counterclockwise rotation R1 of angle π3 from y1 , and y3 by a counterclockwise rotation R2 of angle π3 from y2 . We claim that for any N directions in the plane, we have min

1≤i≤3,1≤s≤N

∠(yi , ls ) ≥

π = α. 6N

The same trick as above shows that the 3N symmetric angle sectors of angle 2α cover the set of all directions and thus the unit circle. Now, the total length of arcs covered by these symmetric cones is 12N α = 2π , and the claim follows. Therefore, the is contained within the intersection of three strips of set of points 2π width w = cos 3n(n−1) , which are orthogonal, respectively, to the directions yj . The intersection is a semiregular hexagon H . Since all three strips have the same width, we find that there are only two different side lengths, as in the figure below:

This means that H is the intersection of two equilateral triangles of the same incenter O. Let A1 , . . . , A6 be the vertices of H . We have then |A1 A2 | = |A3 A4 | = |A5 A6 | = a and |A2 A3 | = |A4 A5 | = |A6 A1 | = b. Moreover, by computing the

156

7 Geometry Solutions

width of H , we find that the lengths a, b are constrained to satisfy a + b = √2 w. 3 Assume from now on that a ≤ b. We divide the hexagon H by means of three segments OP , OQ, OR orthogonal to the three longest sides of the hexagon. The diameter of the pentagon √ OP A1 A2 Q is one of its diagonals, and one computes easily |P Q| = a+ b 2,

|OA1 | =

3b2 +(2a+b)2 √ , 2 3

and |P A2 | ≤ |P Q| since ∠(P A2 Q) ≥ ∠(P QA2 ) = √ 3 2 w,

π 3.

Using the relation between a and b, we obtain that |P Q| ≤ 43 (a + b) ≤ with equality only when a = b and the hexagon is regular. On the other hand, |OA1 | ≤ 2w 3 , with equality when a = 0. This implies that the diameter of each piece is at most √ 3 2 w, where w has the value found above. Comments 63 This version of Borsuk’s problem for finite sets and its higherdimensional generalization for covering by cubes was given in: •

L. Funar: A minimax problem in geometry (in Romanian), Proc. 1986 Nat. Conf. Geometry and Topology - Tîrgoviste, Univ. Bucharest, 91–96, 1988.

Problem 3.38. Prove that any convex body in Rn of unit diameter having a smooth boundary can be partitioned into n + 1 parts of diameter d < 1. Solution 3.38. The first step is to observe that the claim holds for the unit ball B n . We consider the regular spherical simplex obtained by projecting the regular simplex onto the sphere from its incenter. The n + 1 faces of the spherical simplex form a dissection into pieces of diameter smaller than 1, since there are no opposite points lying in the same face. Otherwise, one can compute explicitly the diameter of each piece. Consider the convex body F with smooth boundary W ⊂ Rn . The only point where the smoothness is used is when we assume that there exists a unique tangent hyperplane Tp W at each point p of W . Moreover, there exists a unique unit vector vp in Rn , based at p, that is orthogonal to the tangent hyperplane Tp W and points toward the half-space that does not contain F . This permits us to define a smooth map G : W → S n−1 to the unit sphere centered at the origin O, usually called the Gauss map in differential geometry. We associate to the point p ∈ W the unique point G(p) ∈ S n−1 for which the vectors OG(p) and vp are parallel, and so the tangent space Tp W is parallel to the tangent hyperplane TG(p) S n−1 at the point G(p) on the sphere. Since F is convex and W is smooth, the map G is a smooth diffeomorphism. Consider now the preimage by G of a partition of the unit sphere into pieces of diameter smaller than 1. This yields a partition of W into n + 1 pieces, which we denote by A1 , A2 , . . . , An+1 . We claim that each piece Aj has diameter smaller than 1. Assume the contrary. We can replace the Aj by their closures, without modifying their diameters. Now there should exist two points x, y ∈ Aj at distance 1, which is the diameter of F . Let Tx W and Ty W be the tangent planes at W at x, y respectively. Then Tx W and Ty W are orthogonal to the segment xy. Otherwise, a small perturbation x ∈ W of x in some tangent direction from Tx W , making an obtuse angle with xy, will increase the length of |x y|, but the diameter of W is no larger than 1.

7.2 Combinatorial Geometry

157

Therefore G(x) and G(y) are opposite points on the unit sphere, and moreover, they belong to the same set of the partition, but this is impossible, since the pieces of the partition are of diameter strictly smaller than 1. Comments 64 The question whether any set F in Rn can be partitioned into n + 1 pieces of strictly smaller diameter is known as Borsuk’s problem. Since the diameter of a set equals the diameter of its convex hull, it suffices to consider the problem the convex sets. The result in the problem above was obtained by H. Hadwiger in 1945. The case left open is that in which the convex body has corners. There was no significant progress in the subject until 1993, when J. Kahn and G. Kallai gave an unexpected counterexample by showing that the minimal number f (n) of pieces√needed for a dissection into parts of strictly smaller diameter satisfies f (n) ≥ (1.1) n for large n and that f (n) > n + 1 for n ≥ 1825. The subsets that they considered are finite sets of points. This bound was subsequently lowered by N. Alon and then, by A. Hinrichs, and C. Richter to n = 298. A thorough treatment of the subject is in the recent book of Boltyanski, Martini, and Soltan. However, the 4-dimensional case is still open. • • • •

V. Boltyanski, H. Martini, and P.S. Soltan: Excursions into Combinatorial Geometry, Springer, Universitext, 1997. H. Hadwiger: Überdeckung einer Menge durch Mengen kleineren Durchmesser, Commentarii Math. Helvet. 18 (1945/1946), 73–75, 19 (1946/1947), 71–73. A. Nilli: On Borsuk’s problem, Jerusalem combinatorics ’93, 209–210, Contemp. Math., 178, Amer. Math. Soc., Providence, RI, 1994. J. Kahn and G. Kalai: A counterexample to Borsuk’s conjecture, Bull. Amer. Math. Soc. 29 (1993), 1, 60–62.

Problem 3.39. Let D be a convex body in R3 and let σ (D) = supπ area(π ∩ D), where the supremum is taken over all positions of the variable plane π . Prove that D can be divided into two parts D1 and D2 such that σ (Di ) < σ (D). Solution 3.39. Choose a point in the interior of D and a sequence of chords l1 , l2 , . . . , ln , . . . of D passing through this point, the chords being dense in D. Choose also a sequence ε of positive numbers decreasing rapidly to zero. Let Cn be the set of points at distance less than εn from ln . By choosing ε small enough, we can ensure ∞ −n that ∞σ (Cn ) < 2 σ (D). Set D1 = ∪n=1 Cn and D2 = D − D1 . Then σ (D1 ) < n=1 σ (Cn ) < σ (D). It is also clear that for any plane π , area(π ∩ D2 ) < σ (D). But D2 is a compact and area(π ∩ D2 ) is a continuous function on the plane π . The set of planes that intersect D can be parameterized by a point in D (which will be considered the origin) times the projective space RP 2 , and thus by a compact set. Thus area(π ∩ D2 ) achieves its minimum, which will therefore be strictly smaller than σ (D). Comments 65 The problem was proposed by L. Funar and the solution above was given by C.A. Rogers. A more interesting question is to consider dissections into convex subsets. The additional convexity assumption implies that the boundary structure of

158

7 Geometry Solutions

each piece (at the interior of the body) should be polyhedral. In particular, the minimal number of convex pieces with smaller σ , which the unit ball requires, is 3. If the body D is strictly convex (meaning that the open segment determined by two of its boundary points is contained in the interior of D) or D is a polyhedron without parallel edges, then we can show that four pieces suffice. By a recent result of M. Meyer, two plane sections of maximal area cannot be disjoint. Fix a maximal-area cross-section F and let F (ε) be the set of points of D, at distance less than ε from F . The complement of F (ε) has two components, A+ and A− , which are of strictly smaller σ , because there is no maximal-area plane section contained completely in D − F (ε). Further, F (ε) can be subdivided, by a hyperplane orthogonal to F , into two pieces of smaller σ if ε is small enough. However, we don’t know whether three pieces suffice for an arbitrary convex body. A related problem is whether we have σ (D) ≤ 41 area(∂D), where ∂D denotes the boundary of D. This would imply that the number of maximal area planar sections with disjoint interiors is at most 4, with equality for the regular simplex. • •

L. Funar: Problem E 3094, Amer. Math. Monthly 92 (1985), 427. M. Meyer: Two maximal volume hyperplane sections of a convex body generally intersect, Period. Math. Hungar. 36 (1998), 191–197.

Problem 3.40. If we have k vectors v1 , v2 , . . . , vk in Rn and k ≤ n + 1, then there 1 exist two vectors making an angle θ with cos θ ≥ − k−1 . Equality holds only when the endpoints of the vectors form a regular (k − 1)-simplex. Solution 3.40. Assume that the vectors vj are unit vectors. Then ⎞ ⎛ 2 k k 2 2 |vi − vj | ≤ ⎝ |vi − vj | ⎠ + vi = k |vi |2 = k 2 . 1≤i<j ≤k

1≤i<j ≤k

i=1

i=1

2k Thus there exists at least one pair i, j for which |vi − vj |≤ k−1 . This is equivalent to the fact that the angle θ between vi and vj satisfies the claimed inequality.

Problem 3.41. 1. Consider a finite family of bounded closed convex sets in the plane such that any three members of the family have nonempty intersection. Prove that the intersection of all members of the family is nonempty. " 2. A set of unit diameter in R2 can be covered by a ball of radius

1 3.

Solution 3.41. 1. Assume that the convex sets are A1 , A2 , . . . , Ak and k ≥ 4. We claim that if each k − 1 sets among them have a common point, then all of them should have a common point. The result of the problem follows by using recurrently the previous claim. Before proceeding, we observe that given a set of k ≥ 4 points in the plane, we can find two subsets whose convex hulls intersect nontrivially. Here the convex hull of a set of points is the smallest convex polygon containing these points (and thus all

7.2 Combinatorial Geometry

159

segments pairwise connecting them). In fact, if we choose four points among them, then their convex hull is either a triangle containing the fourth point in its interior, or else a quadrilateral having two diagonals crossing each other. Let pi be a point belonging to ∩j =i,j ≤k Ai and possibly not to Ai . There exists therefore a partition of the set {p1 , p2 , . . . , pk } = P ∪ Q, where P = {pi1 , pi2 , . . . , pim } and Q = {pj1 , pj2 , . . . , pjk−m }, such that the convex hull of the subset P intersects in at least one point the convex hull of the subset Q. Recall that each point of P belongs to ∩i∈{i1 ,i2 ,...,im } Ai , which is convex as the intersection of convex sets. Thus the convex hull of P is equally contained within ∩i∈{i1 ,i2 ,...,im } Ai . The same argument shows that the convex hull of Q is contained in ∩j ∈{j1 ,j2 ,...,jk−m } Aj . Since the two convex hulls have at least one common point, this therefore belongs to ∩ki=1 Ai , as claimed. " 2. It suffices to see that any finite subset is contained in a disk of radius 31 , which " means that the intersection of balls centered at that points of radius 13 is nonempty. By the previous claim, it suffices to consider the case of three points. If the triangle made by these points is obtuse, then we can take the disk whose diameter is the largest side of the triangle. If not, then all angles of the triangle are acute and there exists " at least one angle, say A ≥

π 3.

The circumradius of the triangle is R =

a 2 sin A

≤

1 3.

Comments 66 The first part of the problem is the particular case of the celebrated Helly’s theorem, proved in 1923: if a family of closed convex sets in Rn has the property that any n + 1 among them have a common point, then the intersection of all sets is nonempty. The second claim was generalized by Jung in 1901, who proved that any set of diameter 1 in Rn admits " an unique n-ball of minimal radius covering it, and the minimal radius n is bounded by 2(n+1) . The original proof of Jung’s theorem was quite complicated, but there exists an elementary proof given later by Blumenthal and Wahlin. It is worthy of notice√that any subset of Rn of diameter 1 can be covered also by a , according to Gale. simplex of diameter n(n+1) 2 •

• • •

L.M. Blumenthal, G.E. Wahlin: On the spherical surface of smallest radius enclosing a bounded subset of n-dimensional Euclidean space, Bull. Amer. Math. Soc. 47 (1941), 771–777. L. Danzer, B. Grünbaum, and V. Klee: Helly’s theorem and its relatives, Convexity, Proc. Symposia Pure Math. vol. 7, AMS, 1963, 101–180. D. Gale: On inscribing n-dimensional sets in a regular n-simplex, Proc. Amer. Math. Soc. 4 (1953), 222–225. H.W.E. Jung: Über die kleinste Kugel, die eine räumliche Figur einschliesst, J. Reine Angew. Math. 123 (1901), 241–257.

Problem 3.42. Let T be a right isosceles triangle. Find the disk D such that the difference between the areas of T ∪ D and T ∩ D is minimal.

160

7 Geometry Solutions

Solution 3.42. Let us show that the circle C bounding D has the following two properties, valid for an arbitrary triangle T : 1. The circle C is split into three arcs outside T and three arcs inside T . Then the total length of the outside arcs equals the total length of the inside arcs. 2. The three chords determined by the edges of T have their lengths proportional to the respective side lengths. 1.Assume that C has fixed center. When the radius of C increases by ε, area(T −D) decreases as lin ε 2 , where lin is the length of the inside arcs. On the other hand, area(D − T ) decreases as lou ε 2 , where lou is the length of the outside arcs. Thus we have a local minimum only if (lin − lou) ε 2 does not increase and thus lin = lou . 2. Assume now that C has fixed radius. Since area(T − D) − area(D − T ) = area(T ) − area(D), it is sufficient to minimize area(D − T ). Consider that the center of D is slightly translated by εu, where u is a unit vector. Then we can compute the change of area(S − T ) by summing the contribution of each chord sector. The contribution of A1 A2 is εu × A1 A2 and thus the total variation is εu × A1 A2 + B1 B2 + C1 C2 . The variation must be zero in all directions u and so we obtain the condition A1 A2 + B1 B2 + C1 C2 = 0. This implies that the chords are proportional to the side lengths.

If T is right isosceles triangle of unit side length, then the center O of the circle C belongs to the altitude, by symmetry considerations. Denote by x the distance from the center O to the edges CB and CA, by y the distance from O to the edge AB. The angles determined by the outside √ chords are 2α, 2β, 2β respectively. Since O belongs to the altitude, we have y + 2x = 21 . Moreover, the first claim above implies that √ r 2 − x 2 and α + 2β = π2 and thus cos α = sin 2β. This translates into yr = 2x r2 √ 2 2 2 2 4 2 2 thus √ y r = 4x r − 4x . The second statement above 2implies that r − y / 2 = r 2 − x 2 , and thus introducing above, we obtain 2r = 4x − 1. In particular, x satisfies the equation 16x 4 − 16x 3 − 12x 2 + 8x − 1 = 4x 2 + 2x − 1 4x 2 − 6x + 1 = 0.

7.2 Combinatorial Geometry

161

√

The only convenient root is x = 5−1 4 , since among the remaining ones, one is negative, another one is larger than 1, and the last one would lead to a negative value for r. Thus 0 √ √ 1 √ 1+ 2− 5 r= 5−2 , y = , √ 2 2 2 and the circle is determined by these parameters. Problem 3.43. Let r be the radius of the incircle of an arbitrary triangle lying in the √ 5−1 closed unit square. Prove that r ≤ 4 . Solution 3.43. It is clear that the inradius of a triangle contained in a larger triangle is not bigger than the inradius of the bigger triangle. Thus it suffices to consider only triangles whose vertices lie on the boundary of the square. Further, at least one vertex of the triangle can be assumed to be in a corner of the triangle. If two vertices of the triangle belong to the same edge, then we can assume that these vertices are the endpoints of that edge, as above. Thus it suffices to consider the triangle OAB whose vertices are O = (0, 0), A = (a, 1), and B = (1, b) with 0 ≤ a, b ≤ 1. The inradius of a triangle of area S and semiperimeter p is given by r = pS . Therefore, r=√

1 + a2

+

√

1 − ab . 1 + b2 + (1 − a)2 + (1 − b)2

The inequality to be proved is then equivalent to √ F (a, b) = 1 + a 2 + 1 + b2 + (1 − a)2 + (1 − b)2 − 5 + 1 (1 − ab) ≥ 0 ∂2F > 0 by a simple computation. ∂2a √ 6 √3 2 . Let ε = 1 − (1+ 5)2 1 1. If 1 ≤ b ≤ 4 , set f (x) = F (x, b). Then we observed above that f (x) is √ 1 > 0, then f (x) > 0 for any increasing, and since f (0) = 1 + 5 y − y 2 −2y+2

for 0 ≤ a, b ≤ 1. We have F (a, 0) ≥ 0. Further,

x, and thus f (x) is increasing. In particular, F (a, b) ≥ F (0, b) ≥ 0, provided that 1 4 ≤ b ≤ 1. 2. If 0 ≤ a ≤ 41 and 0 ≤ y ≤ ε, we have

√ x 1−x − f (x) = 1 + 5 b + √ 1 + x2 (1 − x)2 + (1 − b)2 √ x 1−x . ≤ 1+ 5 ε+ √ −√ 2 2 1+x x − 2x + 2

Thus f ( 41 ) < 0. As f is increasing, we have f (x) < 0, for x ∈ 0, 41 , and so f

is decreasing on 0, 41 . But f 41 = F 41 , b = F b, 41 ≥ 0, by the previous case. Thus F (a, b) ≥ 0, for 0 ≤ a ≤ 41 and 0 ≤ y ≤ ε. 3. Since F is symmetric, it suffices to check now the case ε ≤ a ≤ 41 and ε ≤ b ≤ 41 . Actually, we will use only the fact that ab ≥ ε2 . This amounts to saying

162

7 Geometry Solutions

that the area S of the triangle OAB is S = p√2 , 3 3

1−ab 2

≤

1−ε2 2 .

Any triangle satisfies the

obtained from the Heron formula for the area by the means inequality S ≤ inequality. In our case, we have r=

√ S S ≤ √ ≤ 5 + 1. p 3 3S

This settles our claim. Comments 67 The solution presented here is due to Abi-Khuzam and Barbara: • •

L. Funar: Problem 6477, Amer. Math.Monthly 81 (1984), 588. F. Abi-Khuzam, R. Barbara: A sharp inequality and the inradius conjecture, Math. Inequalities and Appl. 4 (2001), 323–326.

Problem 3.44. Let P be a point in the interior of the tetrahedron ABCD, with the property that |P A| + |P B| + |P C| + |P D| is minimal. Prove that AP B = CP D and that these angles have a common bisector. Solution 3.44. Let fA (x) = |AX|. Then fA : E 3 → E 3 is a smooth function on E 3 − {A}. Moreover, the gradient function (∇fA )(X) is the unit vector of direction AX AX, namely |AX| . Consider now the function g = fA + fB + fC + fD for P ∈ int(ABCD). Then g having a minimum at P implies that (∇g)(P ) = (∇fA )(P ) + (∇fB )(P ) + (∇fC )(P ) + (∇fD )(P ) = 0. If a, b, c, d are the unit vectors of directions AP , BP , CP , DP , then the previous conditions reads a + b + c + d = 0, or alternatively, −(a + b) = c + d. Observe now that (a+b) is the direction of the bisector of the angle AP B, while c+d is the direction of the bisector of CP D. Thus these two angles have a common bisector. On the other hand, we have a + b, a + b = c + d, c + d, and because |a| = |b| = |c| = |d|, by simplifying the terms, we obtain that a, b = c, d. This is equivalent to saying that AP B = CP D. Problem 3.45. Let OA1 , . . . , OAn be n linearly independent vectors of lengths a1 , . . . , an . We construct the parallelepiped H having these vectors as sides. Then consider the n altitudes in H as a new set of vectors and further, construct the parallelepiped E associated with the altitudes. If h is the volume of H and e the volume of E, then prove that he = (a1 . . . an )2 . Solution 3.45. Let vi = OAi and wi be the altitude vectors. We have the relations wj , vi = δij vi 2 , where δij is Kronecker’s symbol and , is the scalar product. The linear functionals fi (vj ) = δij vi 2 are linearly independent, and thus the dual vectors wj are linearly independent. Thus E is well defined.

7.2 Combinatorial Geometry

163

Moreover, let A be the matrix consisting of the components of the vi ’s and let B be the matrix consisting of the components of the wj ’s. It follows that AB = diag v1 2 , . . . , vn 2 , where diag(c1 , . . . , cn ) denotes the diagonal matrix with given entries. Therefore vol(E)vol(H ) = | det A| · | det B | = (a1 . . . an )2 . Problem 3.46. Let F be a symmetric convex body in R3 and let AF,λ denote the family of all sets homothetic to F in the ratio λ that have only boundary points in common with F . Set hF (λ) for the greatest integer k such that AF,λ contains k sets with pairwise disjoint interiors. Prove that hF (λ) ≤

(1 + 2λ)3 − 1 . λ3

Solution 3.46. Set Bλ = ∪H ∈AF,λ H . We prove first that Bλ ⊂ (1 + 2λ)F, where αF denotes the homothety of F in ratio α. If v is a point of the boundary ∂F , let v be given by |ov | = (1 + λ)|ov|, where o is the symmetry center of F . Set Fv for the set in AF,λ having center at v . Let x be a point on the boundary of Fv , ox ∩ ∂F = {a}, ov ∩ ∂Fv = {q}, and let vx be parallel to qx, with x ∈ ∂F . Then x = v, xo + v, ox ≥ v, ox, , qv

so that

x. , = qv v, ox ≤ vox

Since F is convex, the intersection of the two segments |oa| ∩ |vx | is nonempty and contains at least one point, say b. Then |vx | ⊂ F , b ∈ F , b ∈ |ox|. Since oa ob ov 1 ≥ = = , ox ox oq (1 + 2λ) the point x belongs to (1 + 2λ)F . Now assume that we have k subsets Fi ∈ AF,λ that have pairwise disjoint interiors. Then their union is contained in Bλ , and all of them are disjoint from F . This means that their union is contained in (1 + 2λ)F − F . Their total volume cannot therefore exceed vol((1 + 2λ)F − vol(F ), and thus their number is bounded by (1 + 2λ)3 − 1 1 vol(1 + 2λ)F − vol(F ) = . vol(F ) λ3 Comments 68 Hadwiger considered the problem of finding estimates for hF (1) (which is called the Hadwiger number of F ) and proved in 1957 that

164

7 Geometry Solutions

n2 + n ≤ hF (1) ≤ 3n − 1 for a convex body F in Rn . Grünbaum conjectured in 1961 (and proved for n = 2) that for any r satisfying these inequalities there exists a convex body F such that hF (1) = r. More generally, the proof used above shows that hF (λ) ≤

(1 + 2λ)n − 1 . λn

We have equality above only if λ1 ∈ Z+ and F is a parallelohedral body. The result is due to V. Boju and L. Funar. There are no good estimates from below except in dimension 2. In fact, if n = 2, the function λhF (λ) approaches an interesting invariant of the oval F , called the intrinsic perimeter. Define the norm ∗ F by means of xyF =

xy , oz

where z is a point of the boundary ∂F such that xy and oz are parallel. This is called the Minkowski norm (or metric) defined by the oval F . The intrinsic perimeter p(F ) of ∂F is the limit of the perimeters of polygons inscribed in F that approach ∂F . For instance, the intrinsic perimeter of a circle is 2π and that of a square is 8. Golab in 1932 and Reshetnyak in 1953 showed that the intrinisc perimeter satisfies the inequalities 6 ≤ p(F ) ≤ 8 with equality in the left side only when F is a hexagon, and on the right side only when F is a rectangle. Then it was proved by V. Boju and L. Funar that the intrinisc perimeter is related to Hadwiger numbers by means of the formula p(F ) = 2 lim λhF (λ), λ→0

and moreover, we have 3+

3 4 ≤ hF (λ) ≤ 4 + , λ λ

with equality on the left side only if λ1 ∈ Z+ and F is an affine regular hexagon. More about packing and covering invariants of convex bodies can be found in the recent book of Böröckzy, and a complete survey on the Minkowski geometry in the book of Thompson. • • •

V. Boju and L. Funar: Generalized Hadwiger numbers for symmetric ovals, Proc. Amer. Math. Soc. 119 (1993), 931–934. K. Böröczky Jr.: Finite packing and covering, Cambridge Tracts in Mathematics, 154, Cambridge University Press, Cambridge, 2004. H. Hadwiger: Über Treffanzahlen be. translationsgleichen Eikörpern, Archiv Math. 8 (1957), 212–213.

7.2 Combinatorial Geometry

•

165

A.C. Thompson: Minkowski geometry, Encyclopedia of Mathematics and Its Applications, 63. Cambridge University Press, Cambridge, 1996.

Problem 3.47. Let denote the square of equations |xi | ≤ 1, i = 1, 2, in the plane, and let A = (a1 |a2 ) be an arbitrary nonsingular 2 × 2 matrix partitioned into two columns. We identify each column with a vector in R2 . Prove that the following inequality holds: a1 , xa2 , x 1 = , minmax 2 A x∈ det A where x, y = x1 x2 + y1 y2 is the usual scalar product. Solution 3.47. Let f (A, x) be the function under minimax. Taking A0 = −11 11 , we obtain 1 max |f (A0 , x)| = . x∈ 2 Thus it suffices to show that for general A, we have max |f (A, x)| ≥ x∈

1 . 2

This is equivalent to showing that | det A| ≤ 1, provided that max |a1 , xa2 , x| = x∈

1 . 2

Assume that the contrary holds, and so | det A| > 1. The set of points in the plane with coordinates (a1 , x, a2 , x) ∈ R2 , for x ranging in the square , is a parallelogram π centered at the origin O and having area 4| det A| > 4. According to our claim,!the parallelogram π is contained in the planar region Z = (u, v) ∈ R2 ; |uv| ≤ 21 . Let us consider the point P = (0, s), where s > 0, is a boundary point of π, and let L ⊂ R2 be the lattice generated by the vectors (1/s, s/2) and (−1/s, s/2). The lattice has area 21 . Then, according to Minkowski’s theorem, the interior of π contains at least one point of the lattice L, different from the origin. Let this point be given by m(1/s, s/2) + n(−1/s, s/2) = ((m − n)/s, (m + n)s/2), where m, n ∈ Z. We can also suppose gcd(n, m) = 1. Since the interior of π is contained in Z, it follows that |m2 − n2 | < 1 and thus m2 = n2 . The only possibilities are (0, ±s) and (±2/s, 0). The points (0, ±s) are on the boundary of π and they are not convenient. If Q = (2/s, 0) ∈ int(π ), then the midpoint S of the segment |P Q| should also belong to the interior of π . However, S = (1/s, s/2) belongs to the boundary curve of equation |uv| = 1/2 of the region Z containing π , so it cannot lie within the interior of π. This contradiction proves our claim. Problem 3.48. We denote by δ(r) the minimal distance between a lattice point and the circle C(O, r) of radius r centered at the origin O of the coordinate system in the plane. Prove that lim δ(r) = 0. r→∞

166

7 Geometry Solutions

Solution 3.48. Let G be a lattice point on the axis Ox such that |OG| = [r]. Draw a perpendicular line at G to Ox that intersects the circle C(O, r) at the point C situated between the lattice points A and B, in the upper half-plane. We choose B to be the point that lies in the interior of the disk D(O, r) of radius r. Notice that |AB| = 1, because these are neighboring lattice points on the same axis line. Let us now draw BD⊥AB, where D is again a lattice point, |BD| = 1, and D is situated outside the disk D(O, r). Since the problem asks for the value at the limit of δ(r) when r → ∞, we can assume that r > 4. Let now consider the line CF that is tangent to the circle C(O, r) and such that F ∈ BD. We show first that |BF | ≤ |BD|. < CDB ≤ ADB = Assume the contrary, namely that |BF | > |BD|. Then CFB π . Furthermore, 4 √ = sin arcsin [r] > r − 1 > 2 = sin π , if r > 4. sin CF B = sin OCG r r 2 4 Thus CF B > π4 , which is false. Consequently, |BF | ≤ |BD| = 1. The triangles CBF and OCG are similar, and therefore √ √ |BC| r 2 − [r]2 |CG| r 2 − (r − 1)2 2r 2 = = < < = √ . |OG| |OG| 2 r r r Now it is clear that

√ 2 |BC| = √ , for r > 4. δ(r) ≤ |BC| ≤ |BF | r

Thus limr→∞ δ(r) = 0. Problem 3.49. Consider a curve C of length l that divides the surface of the unit sphere into two parts of equal areas. Show that l ≥ 2π . Solution 3.49. Let C be the symmetric (opposite, or antipodal) of C with respect to the center O of the sphere. If C ∩ C = ∅, then int C ∩ int C = ∅ and thus the area of the sphere is strictly greater than the sum of areas of int C and int C . By hypothesis, the latter is just the area of the sphere, which is a contradiction. Therefore, C ∩ C contains at least one point, say P . The point P , the symmetric of P with respect to the symmetry center of C ∩C , should also belong to C ∩C , since the later is symmetric. Finally, note that the shortest curve on the sphere that joins two opposite points is an arc of a great circle, and since C is made of two different arcs joining P and P , it follows that l ≥ 2π. Comments 69 More generally, let K be a body with a symmetry center O and set r = minP ∈∂K |OP |, where ∂K denotes the boundary surface of K. If C ⊂ ∂K is a curve that separates two regions on ∂K of the same area, then the length l of C satisfies l ≥ 2π r.

7.2 Combinatorial Geometry

167

Problem 3.50. Let K be a planar closed curve of length 2π . Prove that K can be inscribed in a rectangle of area 4. Solution 3.50. Let bK (θ ) be the width of the figure K in the direction θ , identified with a vector θ ∈ S 1 . If R(K, θ ) denotes the area of the rectangle circumscribing K and having one side parallel to θ , then π . R(K, θ ) = bK (θ )bK θ + 2 Thus

2π 1 1 π 1/2 bK (θ ) 2 bK θ + dθ, θ 2π 0 2 with equality holding iff bK (θ )bK θ + π2 is constant. From the Cauchy–Schwartz inequality, one derives 1

min R(K, θ ) 2 ≤

1 min R(K, θ ) ≤ θ 4π 2

2

2π

bK (θ ) dθ

,

0

the equality holding iff K is a curve of constant width. We are able now to apply the following theorem of Cauchy, which expresses the length L(K) of the curve K in terms of the integral of its width with respect to all directions: 1 2π bK (θ ) dθ. L(K) = 2 0 This implies that 1 L(K)2 ≤ 4, π2 where the second inequality uses the hypothesis L(K) ≤ 2π . min R(K, θ ) ≤ θ

Comments 70 We can consider the parallelograms with a given acute angle φ instead of rectangles. The same argument shows that min Rφ (K, θ ) ≤ θ

cos φ L(K)2 . π2

See also: •

E. Lutwak: On isoperimetric inequalities related to a problem of Moser, Amer. Math. Monthly 86 (1979), 476–477.

Problem 3.51. 1. Consider a family of plane convex sets with area a, perimeter p, and diameter d. If the family covers the area A, then there exists a subfamily with a pairwise disjoint interiors that covers at least area λA, where λ = a+pd+πd 2. 2. Assume that any two members of the family have nonempty intersection. Prove that there exists then a subfamily with pairwise disjoint interiors that covers area at least μA, where μ = πda 2 .

168

7 Geometry Solutions

Solution 3.51. 1. If K is a member of the family and K(d) the set of points at distance less than d from K, then it is known that area(K(d)) = a + pd + π d 2 . Let {K1 , K2 , . . . , Kn } be a maximal subfamily with disjoint interiors. Then every member of the family intersects some Ki , and thus it is contained in some Ki (d). Thus the family is contained within ni=1 Ki (d). Thus the area covered by all members of the family is at most n · area(Ki (d)) ≤ A, while the subfamily above covers na ≥ λA. 2. The diameter of the union of all sets is at most 2d, by the assumption. Its area is therefore at most πd 2 , by the well-known isoperimetric inequality. Thus λA ≤ a, and one single set will cover at least λA. Comments 71 It is still unknown whether the second claim holds without any additional assumption on the sets. Problem 3.52. Let C be a regular polygon with k sides. Prove that for every n there exists a planar set S(n) ⊂ R2 such that any subset consisting of n points of S(n) can be covered by C, but S(n) itself cannot be included in C. Solution 3.52. Let r be the radius of the circle inscribed in C. For n ∈ Z+ , let S(n) be the circle of radius r sec(π/2kn).

1. C cannot cover S(n). In fact, there is no disk of radius greater than r that can be included in C. Here is a proof for this somewhat obvious assertion. For a fixed radius R, let X(R) be the set of centers of circles of radius R included in C. Then X(R) is convex and invariant of rotations centered at the center O of C, of angle 2π/k. Thus X(R) is either empty or contains O. Thus R ≤ r. 2. Let P1 , . . . , Pn ⊂ S(n) be arbitrary points of S(n). We show that C can cover π from O. We place C P1 , . . . , Pn . We fix a point Q in C at a distance of r sec 2kn such that their centers coincide and we rotate C around its center O such that Q runs through S(n). While Q travels along the circle, some of the points Pi will be in the interior of C and the others outside. However, we see that Pi ∈ / C if and only if Q ∈ nj=1 Aj , where each Aj is an π . More precisely, the arc Aj is arc of a circle from outside C, which is of length kn determined by the inequalities πj π πj ≤ P + . i OQ ≤ 2 k k kn n Therefore j =1 Aj is a set of arcs of total length smaller than nkπ/kn · π and thus S − nj=1 Aj has length π, and thus it is nonempty. Now rotate C until Q lies inside S − nj=1 Aj . Then all the Pj belong to C, as claimed. 2

7.2 Combinatorial Geometry

169

Problem 3.53. Let M be a convex polygon and let S1 , . . . , Sn be pairwise disjoint disks situated in the interior of M. Does there exist a partition M = D1 ∪ · · · ∪ Dn such that Di are convex disjoint polygons, each of which contains precisely one disk? Solution 3.53. Let Oi , ri be the center and the radius of the disk Si . For a point X in the plane, let us define hi (X) = XOi2 − ri2 . We consider Dk = {x ∈ M; hk (X) ≤ hi (X), for all i = 2, 3, . . . , n}. We will prove that the Di are convex and Si ⊂ Di . Fix k, for instance k = 1. Define Hi = {X ∈ R2 ; hk (X) ≤ hi (X)}. Then we claim that Hi are half-planes. Let i = 2 for simplicity. Choose the point P ∈ |O1 O2 | such that |P O1 |2 − r12 = |P O2 |2 − r22 . Take for instance |P O1 | = |O1 O2 |2 −(r12 −r22 ) . If l is the line perpendicular to O1 O2 at P and X ∈ l, then we have 2|O1 O2 | |XO1 |2 − r12 = |P O1 |2 + |P X|2 − r12 = |P O2 |2 + |P X|2 − r22 = |XO2 |2 − r22 . If H˜ 2 is the half-plane that contains O1 , then |XO1 |2 − r12 ≤ |XO2 |2 − r22 , for all X ∈ H˜ 2 , |XO1 |2 − r12 ≥ |XO2 |2 − r22 for all X ∈ / H˜ 2 . Therefore, H2 is the half-plane

H˜ 2 . Consequently,

Dk = M

n 1

Hn

i=1,i=k

is a convex set as an intersection of convex sets, and Sk ⊂ Dk . It is clear that Di = M. Problem 3.54. Consider an inscribable n-gon partitioned by means of n − 2 nonintersecting diagonals into n − 2 triangles. Prove that the sum of the radii of the circles inscribed in these triangles does not depend on the particular partition. Solution 3.54. If we have a triangle inscribed in a circle of center O and radius R, of inradius r, then the distances from O to the sides satisfy the well-known formula d ∗ (O, side) = R + r. cyclic

Here d ∗ is the signed distance, which is positive on one half-plane determined by the side and negative on the other half-plane. We write the corresponding formulas for each triangle of the partition and sum up all the terms. Let d1 , . . . , dn be the distances from O to the sides of the n-gon. When looking at the signed distances from O to a diagonal, we observe that each signed distance is considered twice: once with the plus sign and once with the opposite sign. When these terms are summed, they will cancel. We then obtain d1 + · · · + dn = (n − 2)R + r1 + · · · + rn , and thus the sum of inradii r1 + r2 + · · · + rn does not depend on the partition.

170

7 Geometry Solutions

Problem 3.55. Prove that in an ellipse having semiaxes of lengths a and b and total length L, we have L > π(a + b). Solution 3.55. We cut the ellipse along the axes into four congruent pieces. We can further rearrange the four pieces by adding a square of side length b − a as in the figure below:

We want to use the well-known Bonnesen’s isoperimetric inequality, which states that L2 ≥ 4πA for any planar convex domain of perimeter L and area A. The area of the new domain is A + (b − a)2 , where A was the area of the ellipse. Moreover, one knows that the area of the ellipse is A = π ab. Applying this inequality to the domain pictured in the figure above, we obtain L2 > 4π 2 ab + 4π(b − a)2 > π 2 (a + b)2 , whence the claim. Problem 3.56. Let F be a convex planar domain and F denote its image by a homothety of ratio − 21 . Is it true that one can translate F in order for it to be contained in F ? Can the constant 21 be improved? Generalize to n dimensions. Solution 3.56. Let ABC ⊂ F be a triangle of maximal area inscribed in F . Let O denote the geometric centroid of ABC. We use the homothety of center O and ratio − 21 that transforms F into F and ABC into A B C . Since O is the centroid of ABC, we find that A is the midpoint of the segment |BC|, B is the midpoint of |CA|, and C is the midpoint of |AB|.

7.2 Combinatorial Geometry

171

We will show that F ⊂ ABC. Assume the contrary, namely that there exists a point X of F that does not belong to the triangle ABC. By symmetry, it suffices to consider that X belongs to the half-plane determined by the line BC, which does not contain A. Since BC and B C are parallel, it follows that the area of XB C is strictly greater than the area of A B C . Let us consider the inverse image Y of X by the homothety of center O and ratio − 21 . Then Y belongs to F and the area of Y BC is strictly greater than the area of ABC, which contradicts our assumptions. This proves that F ⊂ ABC ⊂ F . The constant 21 is sharp in the case that F is a triangle. The same proof shows that the image F of a convex domain F in Rn by the homothety of ratio − n1 can be translated in order to be contained in F . The constant 1 n is sharp, with equality for the simplex. Problem 3.57. A classical theorem, due to Cauchy, states that a strictly convex polyhedron in R3 whose faces are rigid, must be globally rigid. Here, rigidity means continuous rigidity, in the sense that any continuous deformation of the polyhedron in R3 that keeps the lengths of edges fixed is the restriction of a deformation of rigid Euclidean motions of three-space. Prove that a 3-dimensional cube immersed in Rn remains rigid for all n > 3. Solution 3.57. Three sides are common to a vertex and determine a three-dimensional space R3 ⊂ Rn . The 3-dimensional spaces determined by two adjacent vertices have two sides in common, and therefore they coincide. Thus, the entire cube is immersed in a 3-dimensional subspace of Rn , and therefore it remains rigid. The same argument shows that a rigid polyhedron with the property that exactly three sides meet at any vertex remains rigid after immersion in Rn , for all n > 3. Problem 3.58. Consider finitely many great circles on a sphere such that not all of them pass through the same point. Show that there exists a point situated on exactly two circles. Deduce that if we have a set of n points in the plane, not all of them lying on the same line, then there must exist one line passing through precisely two points of the given set. Solution 3.58. Assume that the contrary holds. Consider the partition of the sphere into (spherical) polygons, induced by the circles, and identify it with a polyhedron. Each vertex will have degree at least 6, since at least three great circles pass through each point. If V is the number of vertices, E the number of edges, and F the number of faces of the spherical polyhedron, then Euler’s formula states that V − E + F = 2. Moreover, by counting the edges in each face and summing over the faces, we obtain 2E = 3F3 + 4F4 + · · · ≥ 3F, where Fk is the number of faces with k sides. Finally, counting the edges adjacent to each vertex and summing over the vertices, we get

172

7 Geometry Solutions

2E = 6V6 + 7V7 + · · · ≥ 6V , where Vk is the number of vertices of degree k, because Vk = 0 when k ≤ 5. Thus 6(E + 2) = 6(V + F ) ≤ 2E + 4E = 6E, which is absurd. Therefore, there exists a vertex of valence smaller than 6 in which exactly two great circles intersect. The second claim can be reformulated using projective duality: if we have n lines in the projective plane, not all passing through the same point, then there exists a point lying on exactly two of the lines. This is equivalent to the problem considered above. Comments 72 The problem was posed by Sylvester in 1839, and it was not settled until 1930, when T. Gallai gave a solution. The solution above is due to N. Steenrod. Moreover, it can be proved that the number of points where exactly two circles intersect is at least 3n/7; see for instance: • •

H.S.M. Coxeter: The classification of zonohedra by means of projective diagrams, J. Math. Pures Appl. (9) 41 (1962), 137–156. G.D. Chakerian, Sylvester problem on collinear points and a relative, Amer. Math. Monthly 77 (1970), 164–167.

Problem 3.59. Given a finite set of points in the plane labeled with +1 or −1, and not all of them collinear, show that there exists a line determined by two points in the set such that all points of the set lying on that line are of the same sign. Solution 3.59. We use projective duality as above and reformulate the question as follows: given n great circles on a sphere, to each one being assigned either +1 or −1, and not all passing through the same point, show that there exists a point lying on at least two circles such that all circles containing that point have the same sign. The circles determine a spherical polyhedron whose edges are labeled by +1 and −1. Assume that the result is not true. Then the edges around each vertex must have both the labels +1 and −1. By symmetry, the number of sign changes in the labels of the edges that one meets when traveling-cyclically around each vertex is at least 4. Denote by N the sum of all such sign changes over all vertices. Then N ≥ 4V . On the other hand, by counting the sign changes as we travel around the faces and summing over all faces, we should again obtain N . Further, the number of sign changes in a k-sided face is even and at most k, so that N ≤ 2F3 + 4F4 + 4F5 + 6F6 + 6F7 + · · · . Moreover, using Euler’s formula and 2E = 3F3 + 4F4 + · · · we obtain 4V − 8 = 4E − 4F = 2(3F3 + 4F4 + 5F5 + · · · ) − 4(F3 + F4 + · · · ) = 2F3 + 4F4 + 6F5 + · · · ≥ N ≥ 4V , which is a contradiction. Thus our claim follows.

7.2 Combinatorial Geometry

173

Comments 73 The result concerning the sign changes is valid more generally when some edges are allowed to be unlabeled, and as such it is known as Cauchy’s lemma. This is an essential ingredient in proving the famous rigidity theorem of Cauchy, which states that a convex polyhedron in R3 whose faces are rigid is globally rigid. Problem 3.60. If Q is a given rectangle and ε > 0, then Q can be covered by the union of a finite collection S of rectangles with sides parallel to those of Q in such a way that the union of every nonoverlapping subcollection of S has area less than ε. Solution 3.60. Observe first that every rectangle Q has a subset H that is the union of a finite collection of rectangles with parallel sides such that: 1. any two members of overlap; 2. area(H ) > δ area(Q), where δ = exp(−1/ε); 3. area(A) < 2ε area(H ), for any A ∈ . If Q = [0, a] × [0, b], we consider rectangles Ai that are parallel and have one vertex at (0, 0) and the other vertex on the curve of equation xy = δ area(Q). Thus each rectangle has area δ. The curved triangle determined by the inequalities xy ≤ δ area(Q), x, y ≥ 0, will be denoted by T . Moreover, we consider a collection of such rectangles such that their union H satisfies area(H ) > 1+ε ε area(T ) = δ area(Q). Further, area(A) = δ area(Q) < ε area(T ). Set Q0 = Q, H0 = H , 0 = and assume that area(Q) = 1, for simplicity. Then Q0 − H0 is a finite union of parallel rectangles Qi,0 and area(Q0 − H0 ) < (1 − δ). Choose in each rectangle Qi a set of type Hi,0 as above. The union of all such Hi,0 and H0 will be denoted by H1 , and the collection of rectangles will be denoted by 1 . Then Q − H1 is also a finite union of parallel rectangles Qi,1 and area(Q − H1 ) < (1 − δ)2 area(Q). We continue this process by defining the sets Hj and the collections j of rectangles such that Q − Hj is a finite union of rectangles Qi,j whose complement has total area area(Q − Hj ) < (1 − δ)j area(Q). We stop this process at the stage n, where n is large enough that (1 − δ)n area(Q) < ε/2. Consider then the union of the collections n obtained so far with the set of rectangles in which Q − Hn decomposes. If we consider a subfamily of disjoint rectangles from , then there is at most one element Ai from any collection i,j covering Hi,j , because all elements in i,j overlap each other. Thus the total area covered by these Ai is at most ε ε area(Hi,j ) < . 2 2 i,j

Since the rectangles from Q − Hn contribute at most 2ε , we are done.

174

7 Geometry Solutions

Problem 3.61. Prove that the 3-dimensional ball cannot be partitioned into three sets of strictly smaller diameter. Solution 3.61. Assume that we can divide the unit ball into three pieces of smaller diameter and let X1 , X2 , X3 be the partition that it induces on the boundary sphere. Thus the diameter of each piece Xi is smaller than 1 − 10h, for some h > 0. We choose a very thin triangulation of the sphere with triangles that have edges smaller than h. Denote by Nj the union of those spherical triangles of the triangulation above that intersect the set Xj nontrivially. It follows that N1 , N2 , N3 are closed subsets of the sphere that are domains, i.e., they are closures of open subsets. In particular, the boundary of each domain Nj is the union of several piecewise linear circles. Moreover, the union N1 ∪ N2 ∪ N3 covers the sphere. Since each triangle has a small size, the diameter of Nj is at most 1 − 8h < 1. In particular, opposite points cannot belong simultaneously to the same set Nj . Let us denote by x ∗ the point of the sphere opposite the point x. We know that the set N1 and its symmetric N1∗ (which is the set of those points x ∗ for which x ∈ N1 ) should be disjoint; otherwise, there would exist in N1 a pair of points at distance 1. Consider the circles determined by the boundary circles of both N1 and N1∗ . The circles associated to Nj are disjoint and distinct from the circles associated to N1∗ , because N1 ∩ N1∗ = ∅. Thus, there is an even number of such circles. Observe that k disjoint circles on the sphere divide the surface of the sphere into k + 1 connected complementary regions. Thus, the collection of boundary circles associated to N1 and N1∗ will split the surface of the sphere into an odd number of connected regions, which we denote by R1 , R2 , . . . , R2m+1 . Consider the symmetric Rj∗ of the region Rj . There are two possibilities: either ∗ Rj is another connected region Rk that is disjoint from Rj , or else Rj∗ = Rj , meaning that Rj is symmetric. The second case happens, for instance, when the region Rj is bounded by two circles, one coming from N1 and the other from N1∗ . If there were no symmetric regions Rj , then the total number of regions would be even, which would contradict our previous claim. Thus there must be at least one connected symmetric region Rj . We consider a point x ∈ Rj and its symmetric x ∗ , which must also belong to Rj , since the region is symmetric. If x ∈ N1 , then all points of the sphere that are connected to x by a path not hitting a boundary circle of N1 must also belong to N1 . Thus, the connected component Rj will be contained in N1 , and so N1 will contain pairs of symmetric points, contradiction. This implies that x ∈ N1 and hence x ∈ N2 ∪ N3 . Let us assume that x ∈ N2 . Then its symmetric x ∗ is not in N2 , since N2 has diameter smaller than 1 and thus x ∗ ∈ N3 . Further, Rj is connected and thus there exists a path γ within Rj that connects x to x ∗ . One knows that N2 and N3 are domains covering the path γ . Moreover, there exists a point z ∈ γ that belongs to both N2 and N3 . In fact, consider the point z ∈ γ closest to x ∗ and that belongs to N2 . There exists such a point since N2 is closed. If z = x ∗ , then both x and x ∗ belong to N2 , contradiction. If z = x ∗ , then the points on the right of z should belong to N3 and thus z belongs to the closure of N3 and thus to N3 .

7.3 Geometric Inequalities

175

Now z ∈ Rj and thus its symmetric z∗ also belongs to Rj . This means that ∈ N1 . But z∗ can belong neither to N2 , since N2 does not contain opposite points, nor to N3 , for the same reason. This contradiction proves the claim. z, z∗

Comments 74 More generally, it is known that the unit n-ball (or equivalently, the (n − 1)-sphere in Rn ) cannot be partitioned into n pieces of strictly smaller diameter. This is related to Borsuk’s problem discussed in comments to Problem 3.38. The result was proved by Lusternik and Shnirelman in 1930 and, independently, by Borsuk in 1932, but the higher-dimensional case n ≥ 4 requires deeper methods from topology, and this opened a new chapter in algebraic topology. The interested reader might consult the survey of Steinlein, which presents both historical remarks and a large number of references. • • •

•

K. Borsuk: Über die Zerlegung einer n-dimensionalen Vollkugel in n-Mengen, Verh. International Math. Kongress Zürich 1932, 192. L.A. Lusternik and L.G. Shnirelman: Topological Methods for Variational Problems, Moscow, 1930. H. Steinlein: Borsuk’s antipodal theorem and its generalizations and applications: a survey, Topological Methods in Nonlinear Analysis, 166–235, Sém. Math. Sup., 95, Presses Univ. Montreal, Montreal, 1985. H. Steinlein: Spheres and symmetry: Borsuk’s antipodal theorem. Topol. Methods Nonlinear Anal. 1 (1993), 1, 15–33.

7.3 Geometric Inequalities Problem 3.62. If a, b, c, r, R are the usual notations in the triangle, show that 1 1 1 1 2 1 ≤ ≤ 2. ≤ 2rR 3 a a2 4r Solution 3.62. We derive from ( a1 − b1 )2 ≥ 0 that 1 1 1 2 1 ≤ . ≤ bc 3 a a2 Moreover, the usual Cauchy–Schwarz inequality shows that (p − a)(p − b) ≤ and its analogues. Then we have 1 2p a+b+c 1 1 1 1 1 2 1 ≤ = = = + + ≤ 2rR abc abc ab bc ca 3 a a2 1 1 p 1 ≤ = = 2. 4 (p − b)(p − c) 4(p − a)(p − b)(p − c) 4r Problem 3.63. If a, b, c are the sides of a triangle, then prove that b2 +c2

ma ka , where ma , wa , ka

2bc ≤ and altitude issued from A.

(b+c)2 4bc

≤

ma wa

c2 4,

and

denotes the respective lengths of the median, bisector,

176

7 Geometry Solutions

Solution 3.63. We have the following formulas computing ma , wa , and ka : √ " bc 1 2 2 2 2 b + c − a , wa = (b + c)2 − a 2 , ma = 2 b+c " bc ka = 2 2 b2 + c2 − a 2 . b + c2 From the triangle inequality, we have a 2 > (b − c)2 , which implies furthermore that (b + c)2 − a 2 < 4bc

(b − c)2 (b − c)2 , ≤ 4bc (b + c)2 − a 2

with equality only if b = c. Adding 1 in both members yields / 2b2 + 2c2 − a 2 b+c 2b2 + 2c2 − a 2 (b + c)2 ≤ ; hence . ≤ √ 4bc (b + c)2 − a 2 (b + c)2 − a 2 2 bc This leads to (b + c)2 b+c ≤ √ 4bc 2 bc

/

2(b2 + c2 ) − a 2 ma − a2 = . 2 (b + c) wa

For the second inequality, it suffices to observe that ma ma b2 + c2 ≥ = . ha ka 2bc Notice that equality in the first case implies b = c, while for the second case either b = c or a 2 = b2 + c2 . Problem 3.64. If S(x, y, z) is the area of a triangle with sides x, y, z, prove that S(a, b, c) + S(a , b , c ) ≤ S(a + a , b + b , c + c ). Solution 3.64. We make the following change of variables: s = (a + b + c)/2, t = s − a, u = s − b, and v = s − c. Using Heron’s formula, the inequality becomes √ √ 4 4 stuv + s t u v ≤ 4 (s + s )(t + t )(u + u )(v + v ). We prove first that for positive x, x , y, y , we have √ xy + x y ≤ (x + x )(y + y ). In fact, this follows from xy + x y + 2 x y · y x ≤ xy + x y + xy + yx , and equality holds when the geometric means of√x y and y x are √ equal. By applying the last inequality to x = st and y = uv twice, we obtain the claim. Moreover, the equality is attained only when t/t = u/u = v/v = s/s , which amounts to a/a = b/b = c/c .

7.3 Geometric Inequalities

177

Problem 3.65. It is known that in any triangle we have the inequalities √ √ 3 3r ≤ p ≤ 2R + (3 3 − 4)r, where p denotes the semiperimeter. Prove that in an obtuse triangle, we have √ (3 + 2 2)r < p < 2R + r. Solution 3.65. We have

cos A−B a+b 2 . = c sin C2 √ √ Since C > π2 , we have sin C2 > 22 and thus c 2 > a + b. Therefore √ √ 2p 2 > (a + b)( 2 + 1) ⇒ 8p 2 √ √ > (3 + 2 2)(a + b)2 > 4(3 + 2 2)ab √ > 8(3 + 2 2)S, where S is the area. This implies that √ √ 3+2 2 S = 3 + 2 2 r. p> p Finally, recall that r = (p − c) tan 2R + r.

C 2

> (p − c) and so p < c + r = 2R sin C + r

2, we have |an | > 2 and consequently an+1 > an for all n ≥ 2. Since the sequence is increasing, it is either convergent or tends to ∞. If we have a finite limit an → λ, then it satisfies λ = λ2 − 2 and thus λ ∈ {−1, 2}, contradicting our assumptions. Therefore, we need |a| ≤ 2 for the convergence. Let α ∈ − π2 , π2 be such that 2 cos α = a. By induction on n, we obtain an = 2 cos 2n α. / Q. Observe that the Assume that α is not commensurable with π, that is, πα ∈ function ϕ(x) = 2x α is continuous. Thus, the map ϕ : R/π Z → R/π Z sends the dense subset {nα (mod π), n ∈ Z} into a dense subset, which is {2n α (mod π ), n ∈ Z}. Therefore, the image under cos, namely {cos 2n α; n ∈ Z}, is dense in [−1, 1], and thus the sequence (an ) cannot be convergent. If α/π = p/q, where p, q ∈ Z, then the sequence (an ) is periodic for n large enough. Thus, in order for the sequence to be convergent it is necessary and sufficient that it be constant for n large enough. This condition reads

204

8 Analysis Solutions

cos 2n α = cos 2n+1 α . Thus 2n α = ±2n+1 α + kπ,

where k ∈ Z,

) ( m π, where m, n ∈ Z . 2n 3 One sees that an is convergent and constant for n ≥ m.

and so

α∈

Problem 4.33. Let 0 < a < 1 and I = (0, a). Find all functions f : I → R satisfying at least one of the conditions below: 1. f is continuous and f (xy) = xf (y) + yf (x). 2. f (xy) = xf (x) + yf (y). Solution 4.33. 1. Let g(x) = f (x) x . Then g(exp(x)) is continuous and additive on I . This implies that g(exp(x)) is linear and hence g(x) = C log x, which implies f (x) = Cx log x for some constant C. 2. Give y the values x, x 2 , x 3 , and x 4 . We find therefore that f x 2 = 2xf (x), f x 3 = xf (x) + x 2 f x 2 = x + 2x 3 f (x), f x 4 = xf (x) + x 3 f x 3 = x + x 4 + 2x 6 f (x), f x 4 = x 2 f (x) + x 2 f x 2 = 4x 3 f (x). From the last two lines, one derives that f (x) = 0 for all but those points x for which x + x 4 + 2x 6 = 4x 3 . This means that f vanishes for all but finitely many (actually 6) values of x. If t ∈ I is such that f (t) = 0, then the first identity shows that f t 2 = 0. n By induction on n we obtain f (t 2 ) = 0. This means that f does not vanish for infinitely many values of the argument, which contradicts our former result. Thus f is identically zero. Problem 4.34. If a, b, c, d ∈ C, ac = 0, prove that −1 + max(|ac|, |ad + bc|, |bd|) ≥ max(|a|, |b|) max(|c|, |d|) 2 Solution 4.34. Set ab−1 = r, dc−1 = s, −1+2 inequality from the statement is equivalent to

√ 5

√ 5

.

= k, k 2 = 1 − k, 0 < k < 1. The

μ = max(1, |r + s|, |rs|) ≥ k max(1, |r|) max(1, |s|) = ν. (i) If |r| ≥ 1, |s| ≥ 1, then μ ≥ |rs| > k|s||r| = ν,

8 Analysis Solutions

205

proving our claim. (ii) If |r| ≤ 1, |s| ≥ 1, then our inequality is equivalent to max(1, |r + s|, |rs|) ≥ k|s|. Moreover, if k|s| ≤ 1 or |r + s| ≥ k|s|, then this is obviously true. Further, let us assume that k|s| > 1 and |r +s| < k|s|. We have then |r|+|r +s| ≥ |s| and consequently, |r| ≥ |s| − |s + r| ≥ |s| − k|s| = (1 − k)|s| = k 2 |s|. Furthermore, |rs| ≥ k 2 |s|2 > k|s| > 1; thus max(1, |r + s|, |rs|) ≥ k|s| holds. (iii) If |r| ≥ 1, |s| ≤ 1, then the inequality is proved as in case (ii), by using the obvious symmetry. Problem 4.35. Let ∞ i=1 xi be a convergent series with decreasing terms x1 ≥ x2 ≥ · · · ≥ xn ≥ · · · > 0 and let P be the set of numbers which can be written in the form i∈J xi for some subset J ∈ Z+ . Prove that P is an interval if and only if xn ≤

∞

xi for every n ∈ Z+ .

i=n+1

Solution 4.35. that xp > ∞ i=p+1 xi and consider 1. Suppose that there exists p such α such that i>p xi < α < xp . Set S(J ) = x . We claim that there does i∈J i such that S(J ) = α. In fact, if J ∩ {1, 2, . . . , p} = ∅, then not exist any J ⊂ Z + x ≥ x > α, because x is decreasing. On the other hand, if J ∩{1, 2, . . . , p} = i p k i∈J ∅, then i∈J xi ≤ i>p xi < α. Finally, P contains numbers smaller than α, for instance i>p xi , as well as numbers bigger than α, such as xp . Thus P is not an interval. 2. Assume now that the hypothesis of the problem ∞ is satisfied and choose any element y from the interval (0, S], where S = i=1 xi . We will show that there exists L such that S(L) = y. Let n1 be the smallest index such that xn1 < y. There exists such a one because limk→∞ xk = 0. By induction, we define nk to be the smallest integer with the property that xn1 + · · · + xnk < y. Let L = {n1 , . . . , nk , . . . }. It is clear that S(L) ≤ y. If p ∈ Z+ \ L, then choose the smallest k with the property that nk > p. Since p does not belong to L, we have xn1 + · · · + xnk−1 + xp ≥ y. This implies that

$∞ % xi + xp ≥ y. i∈L

206

8 Analysis Solutions

(i) Assume that the set Z+ − L is finite. This set is empty only when y = S, which obviously belongs to P . Now take p to be the maximal element of Z+ − L. This implies that all elements bigger than p already belong to L and hence L = {n1 , . . . , nk−1 } ∪ {p + 1, p + 2, . . . }. Then

xi =

i∈L

k−1 i=1

xn1 +

∞

xj ≥

j =p+1

k−1

xni + xp ≥ y,

i=1

from which we obtain that i∈L xi = y. (ii) If Z+ − L is infinite, then for any ε > 0, one can choose some pε ∈ Z+ − L such that xpε < ε. The inequality above for pε implies that xi ≥ y. + i∈L

This is true for any positive ε, and thus

i∈L xi

= y.

Problem 4.36. Does there exist a continuous function f : (0, ∞) → R such that f (x) = 0 if and only if f (2x) = 0? What if we require only that f be continuous at infinitely many points? Solution 4.36. 1. There exists some real number α such that f (α) = 0, and thus f (2α) = 0. Let γ ∈ [α, 2α] be given by γ = sup{x ∈ [α, 2α] such that f (x) = 0}. Since f is continuous, we find that f (γ ) = 0, and thus γ < 2α. Moreover, if xn ∈ R is a sequence such that limn xn = γ , then f (xn ) = 0 for n large enough. In fact, otherwise there exists a subsequence f (xnk ) = 0, which implies that f (2xnk ) = 0, and thus limk f (2xnk ) = 0 = f (2γ ). Finally, recall that γ is the supremum of those x ∈ [α, 2α] for which f (x) = 0 and γ < 2α. Thus for n large such that γ + n1 < 2α, there exists xn with γ < xn < γ + n1 for which f (xn ) = 0. This contradicts the previous claim. Thus there does not exist a continuous f as in the statement. 2. The function ' 0, for 22k < x ≤ 22k+1 , where k ∈ Z, f (x) = 1, for 22k−1 < x < 22k , where k ∈ Z, satisfies the claim and has a countable number of discontinuities. Problem 4.37. Find the smallest number a such that for every real polynomial f (x) of degree two with the property that |f (x)| ≤ 1 for all x ∈ [0, 1], we have |f (1)| ≤ a. Find the analogous number b such that |f (0)| ≤ b. Solution 4.37. Let f (x) = ux 2 + vx + w. We have |f (1)| = |2u + v| = |3f (1) − 4f (1/2) + f (0)| ≤ 3|f (1)| + 4|f (1/2)| + |f (0)| ≤ 3 + 4 + 1 = 8.

8 Analysis Solutions

207

Further, a = 8 because we have equality above for f (x) = 8x 2 − 8x + 1. For the second case, we observe that |f (0)| = |v| = |4f (1/2) − 3f (0) − f (1)| ≤ 8 and thus b = 8 by taking f (x) = 8x 2 − 8x + 1. Problem 4.38. Let f : R → R be a function for which there exists some constant M > 0 satisfying |f (x + y) − f (x) − f (y)| ≤ M, for all x, y ∈ R. Prove that there exists a unique additive function g : R → R such that |f (x) − g(x)| ≤ M, for all x ∈ R. Moreover, if f is continuous, then g is linear. Solution 4.38. Let us show first the existence of the upper limit ϕ(x) = lim supn→∞ f (nx) n . In fact, using the hypothesis, one finds that f (nx) ≤ n(f (x) + M), and hence the sequence f (nx) n is bounded from above, and thus ϕ(x) exists. Furthermore, f (n(x + y)) f (nx) f (ny) M ≤ − , − n n n n which yields ϕ(x) + ϕ(y) = ϕ(x + y); thus ϕ is additive. Next, observe that f (nx) ≤ n(f (x) + M) implies that nx n − f (x) ≤ M, and we can take g(x) = ϕ(x). If f is continuous, then it is easy to see that the lower limit, lim inf n→∞ f (nx) n , coincides with ϕ(x), and moreover, ϕ(x) is continuous. An additive continuous function is therefore linear. In fact, we have f (nx) = nf (x) for n ∈ Z, which yields f (rx) = rf (x) for any r ∈ Q. Since f is continuous and every real number r can be approximated by a sequence of rational numbers, we find that f (rx) = rf (x) for any r ∈ R, and hence the claim. Let us prove the uniqueness of such additive functions. Assume that there exists another additive function h satisfying |f (x) − h(x)| ≤ M, for all x ∈ R. This implies that |g(x) − h(x)| ≤ M, for all x ∈ R. However, an additive function is either zero or unbounded. Therefore, the additive function g − h must vanish.

208

8 Analysis Solutions

Comments 88 Functions f satisfying the condition from the statement are called quasimorphisms. Notice that there exist additive functions ϕ : R → R that are not linear (not continuous of course). Examples can be constructed by choosing a basis of R as a vector space over Q and defining arbitrarily the values of the function on each vector of that basis. Problem 4.39. Show that if f is differentiable and if lim f (t) + f (t) = 1, t→∞

then lim f (t) = 1.

t→∞

Solution 4.39. Let us show first that if limt→∞ (f (t) + αf (t)) = 0, where a = (α) > 0, then limt→∞ f (t) = 0. Let ε > 0 and c < ∞ be such that |f (t) + αf (t)| ≤ aε whenever t ≥ c. Then d αt αt = e (f (t) + αf (t)) ≤ aεeat f (t) e dt for t ≥ c. Therefore, using the mean value theorem we have |eαt f (t) − eac f (c)| ≤ ε eαt − eac , for t > c. This yields

|f (t)| ≤ ea(c−t) · |f (c)| + ε1 − ea(c−t) .

In particular, for sufficiently large t, we have |f (t)| ≤ 2ε, as claimed. Now the statement follows by applying this result to f − 1. Comments 89 Let P (D) be a polynomial in the derivation operator D = write P (D) = (D − λ1 )(D − λ2 ) · · · (D − λn ).

d dt .

Let us

The above result can be generalized as follows. Assume that λi < 0, where denotes the real part; if the function f satisfies limt→∞ P (D)f (t) = 0, then limt→∞ f (t) = 0. This is proved by recurrence on the degree of P . The recurrence step follows from the argument given above. In particular, the claim holds for P (D) = D2 + D + 1. The condition λi < 0 is necessary. In fact, we can consider f (t) = eλi t , which satisfies limt→∞ P (D)f → 0. Also, one notices that P (D) = D0 + D1 + ... + Dn does not fulfill this condition for n ≥ 3. Problem 4.40. Let c be a real number and let f : R → R be a smooth function of class C 3 such that limx→∞ f (x) = c and limx→∞ f (x) = 0. Show that limn→∞ f (x) = limx→∞ f (x) = 0.

8 Analysis Solutions

209

Solution 4.40. According to Taylor’s formula, there exist ξx , ηx ∈ (0, 1) such that 1 f (x + 1) = f (x) + f (x) + f (x) + 2 1 f (x − 1) = f (x) − f (x) + f (x) − 2 Suitable linear combinations of these yield

1 f (x + ξx ), 6 1 f (x + ηx ). 6

1 1 f (x) = f (x + 1) − 2f (x) + f (x − 1) − f (x + ξx ) + f (x − ηx ), 6 6 1 1 2f (x) = f (x + 1) − f (x − 1) − f (x + ξx ) − f (x − ηx ). 6 6 Since x + ξx → ∞ and x − ηx → ∞ as x → ∞, by passing to the limit, one obtains 1 1 lim f (x) = c − 2c + c − · 0 + · 0 = 0, 6 6 1 1 1 lim f (x) = c − c − · 0 − · 0 = 0. x→∞ 2 6 6

x→∞

Comments 90 It can be proved that if limx→∞ f (x) exists and if f (n) is bounded, then limx→∞ f (k) (x) = 0 for any 1 ≤ k < n. See for instance: •

J. Littlewood: The converse of Abel’s theorem, Proc. London Math. Soc. (2) 9 (1910–11), 434–448.

Problem 4.41. Prove that the following integral equation has at most one continuous solution on [0, 1] × [0, 1]: x y f (x, y) = 1 + f (u, v) du dv. 0

0

Solution 4.41. If f1 , f2 are two solutions, then their difference g satisfies x y g(x, y) = g(u, v) du dv. 0

0

Since g is continuous, its absolute value is bounded by some constant M on the square [0, 1] × [0, 1]. Thus x y |g(x, y)| ≤ M du dv = Mxy. 0

0 n n

By induction, we show that |g(x, y)| ≤ M x(n!)y 2 . The induction step follows from

|g(x, y)| ≤ 0

x

0

y

Mun v n M(xy)n+1 du dv = . 2 (n!) ((n + 1)!)2

If we fix x, y, then 0 ≤ |g(x, y)| ≤ lim n→

and thus g(x, y) ≡ 0.

Mx n y n = 0, n!n!

210

8 Analysis Solutions

Comments 91 There exists a continuous solution of this equation, given by f (x, y) = 1 + xy +

√ x2y2 x3y3 + + · · · = J0 −4xy = J0 2i xy , 2!2! 3!3!

where J0 is the Bessel function of order zero. Problem 4.42. Find those λ ∈ R for which the functional equation

1

min(x, y)f (y) dy = λf (x)

0

has a solution f that is nonzero and continuous on the interval [0, 1]. Find these solutions. Solution 4.42. The equation can be written in the form

x

λf (x) =

yf (y) dy + x

0

1

f (y) dy. x

If λ = 0, then f is differentiable, and after derivation, we obtain

1

λf (x) = xf (x) − xf (x) +

f (y)dy =

x

1

f (y)dy. x

Thus f is differentiable and λf (x) = −f (x). This implies that f (x) = A cos μx + B sin μx, where μ = λ−1/2 if λ > 0, and f (x) = A cosh νx + B sinh νx, where ν = (−λ)−1/2 if λ < 0. According to the hypothesis, lim f (x) = 0, n→0

and hence A = 0 for any choice of λ. From the above, we also get lim f (x) = 0,

n→0

and thus Bμ cos μ = 0 and Bν cosh ν = 0. The last equation yields B = 0, and so we do not have nonzero solutions. For λ > 0, if B = 0, then cos μ = 0 and thus 4 μ = (2k + 1) π2 . Thus λ = μ−2 (2k+1) 2 π 2 , where k ∈ Z, and π f (x) = B cos(2k + 1) x, B ∈ R. 2 If λ = 0, the same method gives us f (x) = 0.

8 Analysis Solutions

211

Comments 92 We proved that the integral operator T : C 0 (0, 1) → C 0 (0, 1) defined on the space C 0 (0, 1) of continuous functions on (0, 1) and given by the formula 1 min(x, y)f (y)dy (Tf )(x) = 0

has the eigenvalues

4 (2k+1)2 π 2

and the eigenvectors cos(2k + 1) π2 x.

Problem 4.43. Let X be an unbounded subset of the real numbers R. Prove that the set AX = {t ∈ R; tX is dense modulo 1} is dense in R. Solution 4.43. Let {On }n∈Z+ be a countable base of (0, 1), On = ∅. Let p be the canonical projection p : R → R/Z. Define An = {r ∈ R; p(rX) ∩ On = ∅}. This amounts to An =

x∈X−{0}

1 −1 p ({On }), x

and thus An is open. If F is an interval of length f > 0, then there exists x ∈ X such that xf > 1, which implies that p(xF ) = R/Z and thus there exists r ∈ F with p(rx) ∈ On . This proves that An is dense in R. Since AX =

∞ 1

An ,

n=0

we obtain that AX is also dense in R, as the intersection of open dense subsets. Problem 4.44. Consider P (z) = zn +a1 zn−1 +· · ·+an , where ai ∈ C. If |P (z)| = 1 for all z satisfying |z| = 1, then a1 = · · · = an = 0. Solution 4.44. Consider the polynomial Q(z) = 1 + a1 z + · · · + an zn , which can be written as Q(z) = P (z), when |z| = 1. This implies that |Q(z)| = 1 for all z of unit modulus. Moreover, Q(0) = 1, and so Q has a value in the interior of the disk whose modulus is equal to its maximum on the boundary circle. According to the maximum principle for holomorphic functions, Q has to be constant, Q(z) = 1, whence the claim. Comments 93 W. Blaschke has studied the complex analytic (i.e., holomorphic) functions f : D → C on the unit disk D that are continuous on D and have |f (z)| = 1 for all z satisfying |z| = 1. He has shown that such a function has the form f =σ

n & z − bk , 1 − bk z

k=1

where σ is a constant of modulus one and bk ∈ C are such that |bk | < 1.

212

8 Analysis Solutions

Problem 4.45. Let I ⊂ R be an interval and u, v : I → R smooth functions satisfying the equations u (x) + A(x)u(x) = 0, v (x) + B(x)v(x) = 0, where A, B are continuous on I and A(x) ≥ B(x) for all x ∈ I . Assume that v is not identically zero. If α < β are roots of v, then there exists a root of u that lies within the interval (α, β), unless A(x) = B(x), in which case u and v are proportional for α ≤ x ≤ β. Solution 4.45. The roots of v are isolated and we can suppose that v|(α, β) > 0. Assume that u|(α, β), so that after a possible sign change, u|(α, β) > 0. Also, we have v (α) ≥ 0, v(β) ≤ 0, and since v is not identically zero, then v and v do not have common roots; thus v (α) > 0 and v (β) < 0. We consider w(x) = u(x)v (x) − u (x)v(x). We have w (x) = u(x)v (x) − u (x)v(x) = (A(x) − B(x))uv ≥ 0, and therefore w is increasing; so w(α) ≤ w(β) and thus u(α)v (α) ≤ u(β)v (β). Since u(α) ≥ 0, u(β) ≥ 0, v (α) > 0, v (β) < 0, we obtain u(α) = u(β) = 0, and so w(α) = w(β) = 0. Now w is increasing and hence w ≡ 0 and thus w ≡ 0, which u u yields A(x) = B(x). Moreover, for α < x < β we have vw(x) 2 (x) = ( v ) , and hence v is constant. Comments 94 This result is known as Sturm’s comparison theorem in differential equations. Problem 4.46. Let V be a finite-dimensional real vector space and f : V → R a continuous mapping. For any basis B = {b1 , b2 , . . . , bn } of V , consider the set EB = {z1 b1 + · · · + zn bn , where zi ∈ Z}. Show that if f is bounded on EB for any choice of the basis B, then f is bounded on V . Solution 4.46. We solve first the case n = 1. Given any pair of positive numbers 0 < a < b, there exists some r = r(a, b) (depending on a, b) such that for any x > r, one can find m ∈ Z, with ma ≤ x ≤ mb. In fact, the intervals (ma, mb), for integral m, will cover all of R but a compact set. Thus, for any x > r, one can find d an interval (c, d) containing x, and some m ∈ Z such that a < mc < m < b. Assume that f is unbounded. Let us consider an interval I1 = [a1 , b1 ] with the property that f (x) > 1, for x ∈ I1 . Since f is unbounded, there exist x1 > r(a1 , b1 ) arbitrarily large such that f (x1 ) > 4 and m1 ∈ Z such that m1 a1 < x1 < m1 b1 . Moreover, f is continuous and so f (x) > 2 for any x ∈ [c1 , d1 ], for some small interval around x1 , which can be chosen to be contained in (m1 a1 , m1 b1 ). Set I2 =

8 Analysis Solutions

213

{x ∈ I1 ; m1 x ∈ [c1 , d1 ]}. Thus I2 ⊂ I1 is a compact nonempty interval with the property that f (m1 x) > 2 for any x ∈ I2 . We define in this way a nested sequence of compact intervals I1 ⊂ I2 ⊂ I3 ⊂ · · · Ik ⊂ · · · , and a sequence of integers m1 < m2 < m3 < · · · with the property that f (x) > 2k for x ∈ mk Ik . This nested sequence of nontrivial compact intervals must have nontrivial intersection. Let a ∈ ∩∞ k=1 Ik . Then the restriction of f to aZ is unbounded, since for any k, there exists some mk with f (mk a) > 2k , by making use of the fact that a ∈ Ik . Let now deal with the general case. Let zk ∈ V be a sequence for which f (zk ) is f is continuous. unbounded, e.g., f (zk ) > 2k+1 . Then zk is unbounded, because Let B = {b1 , . . . , bn } be a basis of B and write zk = ni=1 cik bi in the basis B. There exists at least one i0 ∈ {1, 2, . . . , n} such that the sequence ci0 k is unbounded; otherwise, zk would be bounded. One can slightly modify the basis B so that this condition is satisfied for all i. In fact, assume that cik is unbounded for 1 ≤ i ≤ m < n and cik is bounded for m + 1 ≤ i ≤ n. Take then the new basis vectors n

b1 = b1 −

bi , bi = bi , when i ≥ 2.

i=m+1

In the new basis, one can express the vector zk as zk =

m i=1

cik bi +

n

(cik + c1k )bi ,

i=m+1

and all components are now unbounded. k = {x; x = nWe consider now the intervals Jik with the property that U k i=1 αik bi , with αik ∈ Jik } is a neighborhood of zk and f (x) > 2 , for all x ∈ Uk . The argument for n = 1 shows that there exists a sequence of integers (mik ) going to infinity such that we have a nested sequence of nontrivial compact intervals ··· ⊂ Take

1 1 Jik+1 ⊂ Jik ⊂ · · · ⊂ Ji1 . mik+1 mik ∞ 1 1 γi ∈ Jik mik k=1

+ = {γi bi , i ∈ {1, 2, . . . , n}}. It follows that f is unbounded and construct a new basis B on EB+. Problem 4.47. It is known that if f, g : C → C are entire functions without common zeros then there exist entire functions a, b : C → C such that a(z)f (z)+b(z)g(z) = 1 for all z ∈ C. 1. Prove that we can choose a(z) without any zeros. 2. Is it possible to choose both a and b without zeros?

214

8 Analysis Solutions

Solution 4.47. 1. We know that Af + Bg = 1 for some entire functions A and B. Therefore, for any entire function λ, we have also (A + λg)f + (B − λf )g = 1. We now choose a = A+λg such that A+λg = eh . If z0 is a zero of g, we need to have eh(z0 ) = A(z0 ) (with the same multiplicity). This is possible from the interpolation theorem for entire functions, because A(z0 )f (z0 ) = 1. 2. Let f, g ∈ C[z] be nonconstant polynomials of different degrees. We will show that there are no entire functions a, b such that af + bg = 1 and a, b have no zeros. There are no integer nonconstant functions. First, there exist nonzero polynomials A and B such that Af + Bg = 1. As above, for any entire function λ : C → C, one can construct other solutions, namely a = A − λg and b = B + λf . Let us now show that any pair a, b can be obtained in this way. We have (A − a)f + (B − b)g = 0. Thus the set of zeros of (A − a) contains the set of zeros of g (since f and g have no common zeros). In particular, there exists an entire function λ such that A − a = λg. In a similar way, there exists an entire function μ such that B − b = μf . The condition above shows that λ = μ, as claimed. If a, b do not have zeros, then the meromorphic function F = λ1 must be distinct everywhere from the rational functions a1 , a2 , a3 given below: a1 (z) = 0,

a2 (z) = g(z)/A(z),

a3 (z) = −f (z)/B(z).

Consider the meromorphic function G=

F − a1 a3 − a2 · . F − a 2 a3 − a 1

The points where G(z) ∈ {0, 1, ∞} are among those satisfying one of the conditions a1 (z) = a2 (z), a1 (z) = a3 (z), a2 (z) = a3 (z), and therefore they are finitely many. According to Picard’s theorem, G does not have an essential singularity at ∞ and thus G is rational. This implies that F is rational; hence λf is a polynomial and thus a and b are polynomials. But polynomials without zeros are constant, and this is impossible since f and g have different degrees. Problem 4.48. Consider a compact set X ⊂ R. Show that a necessary and sufficient condition for the existence of a monic nonconstant polynomial with real coefficients h ∈ R[x] such that |h(x)| < 1 for all x ∈ X is the existence of monic nonconstant polynomial g(x) ∈ R[x] such that |g(x)| < 2 for all x ∈ X. Prove that 2 is the maximal number with this property. Solution 4.48. Let P(X) be the Banach space of real polynomials endowed with the norm θ = supx∈X |θ (x)|. We define the nonlinear operator A : P(x) → P(x) given by

8 Analysis Solutions

215

1 (Af (x)) = f 2 (x) − f 2 , for f ∈ P(X). 2 If f is monic, then Af is also monic and we have Af = 21 f 2 . An easy induction implies that 2f 2n An f = . n 22 Moreover, if f < 2, then there exists some n for which An f < 1. This proves the first part. Let X = [−2, 2]. The norm of p(x) = x on X is p = 2. We will prove that any monic polynomial on X has norm at least 2. Assume the contrary, namely that there exists f ∈ P(x), monic, with f < 2. We have two intermediate results: 1. There exists an operator T : P(X) → P(X) that takes monic polynomials of degree 2k to monic polynomials of degree k such that T P < P . 2. Assuming the existence of a monic polynomial of norm f < 2, there exists n ∈ Z+ and a monic polynomial g, of degree 2n , with g < 2. By points 1 and 2 above there exists a monic polynomial g of degree 2n with g < 2 on [−2, 2], which implies that there exists a monic polynomial of degree 1 such that g < 2 on [−2, 2], which is obviously false. ˜ Proof of claim 1. If h ∈ P(x), let us consider h(x) = 21 (h(x) + h(−x)), which ˜ is even and thus can be written as a polynomial in x 2 . Thus h(x) = Q x 2 , where now Q : [0, 4] → R. Define T (h)(x) = Q(x + 2). It is clear that T has the desired properties. Proof of claim 2. If f has degree 2m p, where p is odd, then t = T m f is monic of degree p, t < 2. There exists n ∈ Z+ such that p divides 2n −1, and so 2n −1 = pq. We define g ∈ P(x) by means of g = t (x)q x. It is immediate that g < 2, and the proof is complete.

9 Glossary

9.1 Compendium of Triangle Basic Terminology and Formulas 9.1.1 Lengths in a Triangle We used the standard notation for the important features of a triangle, as follows: • • • • • • • • • •

a, b, c the sides lengths A, B, C the angles ma , mb , mc the medians ha , hb , hc the altitudes wa , wb , wc the angle bisectors R the radius of the circumcircle r the radius of the incircle ra , rb , rc the radii of the extrinsic circles S the area p the semiperimeter

We collect below some useful identities between these quantities: • • • • • • • • • • •

S = 21 bc sin A = (p(p − a)(p − b)(p − c))1/2 = 4Rr cos A2 cos B2 cos C2 , S = rp, 4RS = abc, ra (p − a) = rp = S, ra = p tan A2 , 4R + r = ra + rb + rc , aha = 2S, 3 2S 2 ha hb hc = 8S abc = R , 1 1 1 1 1 1 1 ha + hb + hc = ra + rb + rc = r , m2a + m2b + m2c = 43 a 2 + b2 + c2 , √ √ bc p(p − a), wa = 2 b+c " p(p−b)(p−c) , ra = p−a

218

9 Glossary

•

2 2 a 2 + b2 + " c = 8R (1 + cos A cos B cos C),

•

sin

• •

cos A + cos B + cos C = 1 + 4 sin " tan A2 = (p−b)(p−c) p(p−a) ,

• •

a b c R = 2 sin A = 2 sin B = 2 sin C , 2 2 2 2 4ma = 2b + 2c − a .

A 2

=

(p−b)(p−c) , bc

A 2

sin

B 2

sin

C 2

=1+

r R,

9.1.2 Important Points and Lines in a Triangle • • • • •

•

•

•

•

• •

•

•

The circumcenter O is the center of the triangle’s circumcircle. It can be found as the intersection of the perpendicular bisectors of the sides. The incenter I is the center of the incircle and thus the intersection of the angle bisectors. The orthocenter H is the intersection of the three altitudes of the triangle. The centroid G is the intersection of the three medians. outside the The excenter JA is the center of the excircle lying in the angle BAC, triangle ABC and tangent to the lines AB, AC and to the segment BC. There are three excenters, JA , JB , JC , using similar definitions. If MA , MB , MC are the midpoints of BC, CA, AB respectively, then the three lines MA JA , MB JB , MC JC are concurrent, and their intersection point is called the mittenpunkt. If SA , SB , SC are the contact points of the excircles with BC, CA, AB respectively, then ASA , BSB , CSC are concurrent and their intersection is called the Nagel point N a. Let TA , TB , TC be the contact points of the incircle with sides BC, CA, AB respectively. Then the lines ATA , BTB , CTC are concurrent, and their intersection point is called the Gergonne point Ge. The isogonal conjugate of the point X is the intersection of the three concurrent lines that are the reflections of AX, BX, CX about the angle bisectors of A, B, C respectively. The intersection point of the symmedians is called the symmedian point (or the Lemoine point). It is the isogonal conjugate of the centroid G. Two lines passing through the vertex A of a triangle ABC are called isotomic if they cut the side BC in points symmetric with respect to the midpoint of BC. Given a point P , the isotomic lines of the Cevians AP , BP , and CP meet at a point P called the isotomic point of P . The Fermat point F of an acute triangle is the point that minimizes the sum of the distances to the vertices. If we draw equilateral triangles BCA , ABC , ACB on the outside of the triangle ABC, then the three lines AA , BB , CC are concurrent at F . The Euler line: the points O, G, H are collinear and |H G| : |GO| = 2 : 1.

9.1 Compendium of Triangle Basic Terminology and Formulas

• •

219

The Gergonne point Ge, the triangle centroid G, and the mittenpunkt M are collinear, with |Ge G| : |GM| = 2 : 1. The Nagel line: the points I, G, N a are collinear and |N a G| : |GI | = 2 : 1.

Kimberling tabulated and enumerated properties of more than 1477 centers in a triangle (later Brisse extended the list to 2001 items). The interested reader might consult his updated web page: •

C. Kimberling: see http://faculty.evansville.edu/ck6/encyclopedia/ETC.html

9.1.3 Coordinates in a Triangle We are given a reference triangle ABC. Barycentric coordinates. The barycentric coordinates of the point P are given by an ordered triple FP = (tA , tB , tC ) (up to nonzero scalar multiplication) such that P is the centroid of the system consisting of the three masses tA , tB , tC located at the respective vertices A, B, C of the triangle, i.e., the mass tA is located at A, etc. These coordinates were introduced by Möbius in 1827. The coordinates tA are proportional to the areas area(P BC). If they are actually equal to these areas, they are called homogeneous barycentric coordinates. Here are the coordinates of the principal points in a triangle. • • • • • • •

FG = (1, 1, 1) FO = a 2 b2 + c2 − a 2 , b2 a 2 + c2 − b2 , c2 a 2 + b2 − c2 FJA = (−a, b, c), FI = (a, b, c) FGe = ((p − b)(p − c), (p − c)(p − a), (p − a)(p − b)) FNa = (p− a, p − b, p −c) a 2 + c2 − b2 a 2 + b2 − c2 , a 2 + b2 − c2 b2 + c2 − a 2 , FH = 2 a + c2 − b2 b2 + c2 − a 2 FK = a 2 , b 2 , c 2

The line determined by the points (s1 , s2 , s2 ) and (t1 , t2 , t3 ) has the equation ⎛ ⎞ s 1 s2 s3 det ⎝ t1 t2 t3 ⎠ = 0. x1 x2 x3 Trilinear coordinates. The trilinear coordinates of a point P are given by an ordered triple of numbers proportional to the directed distance from P to the sides, up to multiplication by a nonzero scalar. If we normalize them so that they give the distances to the sides, then these are called the exact trilinear coordinates. Trilinear coordinates are also known as homogeneous coordinates. They were introduced by Plücker in 1835. For instance, the vertices have trilinear coordinates (1 : 0 : 0), (0 : 1 : 0), and (0 : 0 : 1).

220

9 Glossary

For most important points in a triangle, the trilinear coordinates are given by triples of the form (f (a, b, c) : f (b, c, a) : f (c, a, b)), for some function f on the sides a, b, c. For instance, we have fH (a, b, c) = cos B cos C, fG (a, b, c) = a1 , fO (a, b, c) = cos A, fI (a, b, c) = 1. For the Fermat point, we can take fF (a, b, c) = sin(A + π3 ). The homogeneous barycentric coordinates corresponding to (x : y : z) are (ax, 1 1by,1 cz). The trilinear coordinates of the isogonal conjugate of (x : y : z) are x, y, z .

9.2 Appendix: Pell’s Equation Pell’s equation is the Diophantine equation x 2 − dy 2 = 1. The first treatment of this equation was actually given by Lord William Brouncker in 1657, but his solution was erroneously ascribed to John Pell by Euler. Lagrange developed the theory of continued fractions, which gives the actual method of finding the minimal solution, and published the first proof in 1766. However, some methods (the cyclic method) were already known to the Indian mathematicians Brahmagupta and Bhaskara in the twelfth century. General solutions. Pell’s equation has infinitely many solutions if d is not a perfect square and none otherwise. If (x0 , y0 ) denotes its minimal solution (called also the fundamental solution) different from (1, 0), then all solutions (xn , yn ) are obtained from it by means of the recurrence relations xn+1 = x0 xn + dy0 yn ,

yn+1 = y0 xn + x0 yn ,

x1 = x0 , y1 = y0 .

The fundamental solution. Thus the main point is finding the fundamental solution. This can be achieved using continued fractions. According to Lagrange, the continued fraction of a quadratic irrational number (i.e., a real number satisfying a quadratic equation with rational coefficients that is not rational itself) is periodic after some point; in our case, even more can be obtained, namely

√ d = a0 , a1 , . . . , am , 2a0 . The numbers aj can be calculated recursively, as follows. Set a0 = the sequences Q0 = 1, P0 = 0, P 1 = a0 , Q1 = d − a02 , Pn = an−1 Qn−1 − Pn−1 , Qn = Then

a0 + Pn an = . Qn

d−Pn2 Qn−1 .

√ d and construct

9.2 Appendix: Pell’s Equation

221

Furthermore, consider the sequences p0 = a0 , q0 = 1, p1 = a0 a1 + 1, q1 = a1 , pn = an pn−1 + pn−2 , qn = an qn−1 + qn−2 . Then we have the identities pn2 − dqn2 = (−1)n+1 Qn+1 , which permit one to construct solutions to the Pell-type equations. If am+1 = 2a0 , as happens for a quadratic irrational, then Qm+1 = 1. In particular, we find that the fundamental solution is determined by the parity of m, as follows: x0 = pm , y0 = qm , if m is odd, x0 = p2m+1 , y0 = q2m+1 , if m is even. Observe that pr /qr = [a0 , a1 , . . . , ar ]. The general solution (xn , yn ) is obtained by means of the trick x 2 − dy2 = (x02 − dy02 )n = 1, and setting x+

√ √ n dy = x0 + dy0 ,

we obtain the family of solutions √ √ n n x0 + dy0 + x0 − dy0 , xn = 2

x−

√

√ n dy = x0 − dy0 ,

yn =

x0 +

√

dy0

n

√ n − x0 − dy0 . 2

The negative Pell equation. The Pell-like equation x 2 − dy 2 = −1 can also be solved, but it does not always have solutions. When it has one, then it has infinitely many. A necessary condition for this equation to be solvable is that all odd prime factors of d be of the form 4k + 1, and that d ≡ 0 (mod 4). However, these conditions are not sufficient, as can be seen from the equation for d = 34, which has no solutions. The method of continued fractions works for these equations as well. In fact, for this Pell-type equation, we have the fundamental solution x0 = pm ,

y0 = qm ,

if m is even.

The equation does not have solutions if m is odd. Furthermore, the general solution is given again by √ √ √ √ n n n n x0 + dy0 + x0 − dy0 x0 + dy0 − x0 − dy0 , yn = , xn = 2 2

222

9 Glossary

but only for odd n. Pell-like equations. The Pell-like equation x 2 − dy 2 = c √ can be solved using the same ideas. If |c| < d, then the equation has solutions if and only if ' 8 c ∈ Q0 , −Q1 , . . . , (−1)m+1 Qm+1 . √ Moreover, if c > d, then the procedure is significantly more complicated. See for instance the comments of Dickson on this subject. It is clear that using the solutions (xn , yn ) to the companion Pell equation x 2 − dy 2 = 1 and a particular solution (u, v) of the general Pell-type equation from above, we are able to find infinitely many solutions by means of the recurrence un = xn u ± dyn v,

vn = xn v ± yn u.

However, even if we start with the minimal solution, we cannot always obtain all solutions of the equation using this recurrence. The reason is that we might well have several fundamental solutions. For instance, if d = 10 and c = 9, then we have the fundamental solutions (7, 2), (13, 4) and (57, 18). The general equation ax 2 − by 2 = c can be reduced to the former equation by setting d = ab, x = ax, c = ac and looking for those solutions where x is divisible by a. The degree-two equation in two unknowns. Moreover, the general degree-two equation ax 2 + bxy + cy 2 + dx + cy + f = 0 where a, b, c, d, e, f ∈ Z, can be reduced to either Pell-type equations or elliptic equations. 1. The case = b2 − 4ac < 0. The equation represents an ellipse in the plane, and therefore it has only finitely many integer solutions. These equations were solved by Gauss and earlier considered by Euler. According to Dickson one can find all its solutions using the solutions to Pell’s equations. 2. If = 0, the parabolic-type equation becomes 1 (2ax + by + d)2 + 2(ae − bd)y + 4af − d 2 = 0. 4a If 2ae − bd = 0, then the equation is equivalent, through a linear transformation, to (2ax + by + d)2 = d 2 − 4af ; hence the parabola degenerates into two lines. If d 2 − 4af isa perfect square, then we have two infinite families of solutions 2ax + by + d = ± d 2 − 4af ; otherwise, there are no integer solutions.

9.2 Appendix: Pell’s Equation

223

If 2ae−bd = 0, then we have to solve the equation 2(ae−bd)y = −t 2 +d 2 −4af , for an arbitrary parameter t ∈ Z. This depends only on computing the residues modulo 2(ae − bd) of the perfect squares, and yields infinitely many solutions if one exists. Furthermore, 2ax = t − d − by determines x. 3. If > 0, then the hyperbolic-type equation becomes: a(x − 2cd + bc)2 + b(x − 2cd + bc)(y + 2ac + bd) + c(y − 2ac + bd)2 = − ac2 + cd 2 + ef 2 − bdc − 4acf , which, through a transformation u = x − 2cd + bc, v = y − 2ac + bd, becomes au2 + buv + cv 2 = h. This equation can be reduced to a Pell equation. It will have then either zero or infinitely many solutions. The Ankeny–Artin–Chowla Conjecture. Although the Pell equation seems to be well understood, there are many subtle questions concerning its solutions. Here is a problem that has resisted to all efforts until now. 1. If p is a prime p ≡ 1 (mod 4) and y0 is the smallest positive value of y such that x 2 − py 2 = −4, then y0 ≡ 0 (mod p). 2. If p is a prime p ≡ 3 (mod 4) and y0 is the smallest positive value of y such that x 2 − py 2 = 1, then y0 ≡ 0 (mod p). The first conjecture is due to Ankeny, Artin, and Chowla, and the second to Mordell. The Thue theorem for higher degree. The situation of equations of higher degree is completely different. Let f (x, y) = a0 x n + a1 x n−1 y + a2 x n−2 y 2 + · · · + an y n be a homogeneous form of degree n. If n ≥ 3 and f is irreducible over Q, then the equation f (x, y) = g(x, y), where g(x, y) is an arbitrary form of degree n − 2 (not necessarily homogeneous), has only finitely many solutions. This is a celebrated theorem due to Thue and based on results of Siegel and Roth. Later A. Baker gave explicit bounds on the size of the solutions for a wide class of such f . •

•

L.E. Dickson: Pell equation: ax 2 + by 2 + c made square, Chapter 12 in History of the Theory of Numbers, Vol. 2: Diophantine Analysis, New York, Chelsea, 341–400, 1952. H.W. Lenstra, Jr.: Solving the Pell equation, Notices Amer. Math. Soc. 49 (2002), 2, 182–192.

Index of Mathematical Results

ˇ Cebotarev density theorem, 83 Ankeny–Artin–Chowla conjecture, 223 Birch–Swinnerton-Dyer conjecture, 81 Blaschke theorem, 211 Bonnesen’s isoperimetric inequality, 170 Bonsé–Pósa inequality, 46 Borsuk problem, 153, 156, 175 Borsuk–Lusternik–Shnirelman theorem, 175 Catalan’s conjecture, 56 Cauchy’s integral formula, 167 Cauchy’s rigidity theorem, 171, 173 Cauchy–Schwartz inequality, 167 Cauchy–Schwarz inequality, 175 Ceva’s theorem, 139 Chebyshev theorem, 40, 45, 46, 58 Chinese Remainder theorem, 82 Chinese remainder theorem, 50, 78 congruent numbers problem, 80 De Moivre’s identity, 60 Dirac theorem, 97 Dirichlet theorem on arithmetic progressions, 50, 82, 97 Eisenstein criterion, 38 Erd˝os conjecture, 114 Erd˝os theorem, 126 Erd˝os–Straus conjecture, 63 Euler conjecture, 42 Euler criterion, 72, 73 Euler theorem in a triangle, 182

Fermat equation, 42, 67 four color theorem, 104 Frobenius coin problem, 50 fundamental theorem of symmetric polynomials, 85, 89 Gowers theorem, 123 Green–Tao theorem, 123 Hadamard inequality, 186 Hahn–Banach lemma, 131 Heawood conjecture, 103 Helly’s theorem, 159 intermediate value theorem, 188 isoperimetric inequality, 147 Jensen inequality, 179 Jung’s theorem, 159 Kahn–Kallai theorem, 156 Kaprekar constant, 95 Lagrange four squares theorem, 74 Lagrange mean value theorem, 194 Lagrange multiplier method, 189 Lagrange theorem on periodic continued fractions, 220 Laplace method, 198 Lehmer conjecture, 57 Ljunggren equation, 69 Maclaurin series, 189

226

Index of Mathematical Results

maximum principle for holomorphic functions, 211 Means inequality, 161, 179, 180, 189 Millennium problems, 81, 195 Minkowski’s theorem, 165

Roché inequality, 28, 182 Sturm comparison theorem, 212 superperfect numbers conjecture, 60 Sylvester’s problem, 172 Szemerédi theorem, 122

odd perfect number conjecture, 59 Pell equation, 43 Pell’s equation, 220 Picard theorem, 214 pigeonhole principle, 86, 92, 96, 97, 108, 112

Taylor formula, 209 Thales’ theorem, 140 Thue’s theorem, 223 Ulam problem, 126 Van der Waerden theorem, 122

quadratic reciprocity theorem, 50 Rayleigh–Beatty theorem, 101 Riemann hypothesis, 195

Waring problem, 52, 74, 76 Wiles theorem, 42 Wilson theorem, 44, 53

Index of Mathematical Terms

abundant number, 9, 81 altitude, 22, 25, 27, 139, 160, 162, 175, 217 annulus, 190 area, 8, 19, 23, 25–27, 70, 80, 124, 140–142, 145, 154, 157, 159, 161, 165–167, 170, 173, 176, 177, 181, 182, 190, 217 arithmetic progression, 4, 7, 12, 40, 50, 56, 82, 87, 97, 122 asymptotic, 16, 78, 106, 114, 128 ball, 25, 27, 143, 156–159, 174, 175 Banach space, 214 barycentric coordinates, 142, 219 Bernoulli numbers, 54 Bessel function, 210 binomial binomial coefficient, 14, 41, 62, 97, 98, 105, 109 binomial expansion, 38, 62, 104 binomial coefficient, 3, 37 bisector, 21, 25, 27, 133, 134, 162, 175, 217 body, 25, 145, 156–158, 166 boundary, 18, 25, 150, 151, 156–158, 161, 165, 166, 174, 211 cardinal of a set, 15, 17, 50, 78, 88, 92, 100–102, 104, 108, 111, 112, 117, 128, 191 cardinality of a set, 15 centroid, 218 Cevian, 139 choice number, 3, 37 chord, 21, 22, 24, 135–137, 152, 157, 159, 160

circular permutation, 62, 190 circumcenter, 218 circumscribable, 23, 140 collinear points, 23, 126, 138, 140–142, 172, 182, 218, 219 commensurable, 126, 203 compact, 34, 151, 157, 212–214 complex number, 13, 24, 30, 89, 92, 126, 135, 136, 147, 148, 190, 200 complex plane, 55 computer search, 43, 63, 65 computer-assisted proof, 104 concave, 149 congruence modulo p, 3, 9, 39, 40, 44, 50, 52, 61, 63, 66, 72, 73, 78–80, 82, 83 congruence transformation, 24, 151 congruent number, 80 continued fraction, 220 convergent, 9, 30–33, 77, 78, 190, 191, 193, 194, 203, 205 convex, 16, 22–26, 106, 136, 139, 143, 144, 147, 149, 156–158, 167, 168, 170, 171 convex hull, 136, 156, 158 covering, 27, 173 critical, extremal point, 180, 190 deficient number, 9, 81 dense, 34, 126, 131, 142, 157, 203, 211 density, 83, 109, 122, 128 determinant, 11, 86, 87, 109, 138, 141, 186 diameter, 24, 25, 145, 147, 151, 152, 154, 156, 158, 167, 174 differential equation, 198 digits, 4, 5, 13, 40, 47, 95

228

Index of Mathematical Terms

Diophantine equation, 42, 53, 56, 64, 67, 69, 70, 80, 115, 220 discriminant, 48, 83, 194 distance, 16, 19, 20, 22–24, 26, 107, 122, 126, 128, 130, 131, 136, 142, 143, 145, 146, 148, 151, 152, 165, 167, 169, 219 domain, 26, 170, 171, 174 eigenvalue, 90, 91, 211 Euler inequality, 28, 177 Euler line, 22, 138, 218 Euler totient function, 3, 6, 7, 39, 51, 57, 61, 92 Euler–Poincaré characteristic, 103, 150, 171, 172 Eulerian circuit, 19, 123 excenter, 218 factorial, 3, 37 Fermat point, 22, 138, 218 Fibonacci sequence, 118, 185 fractional part, 193 function, 12, 23, 31, 33, 88, 144, 165, 194, 204, 207 additive function, 208 bijective function, 15, 102 concave (or convex) function, 179 continuous function, 33, 34, 71, 153, 157, 206, 209, 210 decreasing function, 31, 139, 196 entire function, 34, 213 function, 104 generating function, 16, 106 holomorphic function, 211 hyperbolic trigonometric function, 203 increasing function, 15, 64, 101, 110, 189 meromorphic function, 214 periodic function, 31, 193 positive function, 32, 198 rational function, 13, 42, 92 smooth function, 32–34, 87, 162, 198, 199, 208, 212 Gauss map, 156 Gergonne point, 23, 141, 218, 219 graph, 15, 19, 97, 102, 103, 123, 147, 150 Hadwiger number, 163 Hamiltonian cycle, 97

homographic division, 138 homothety, 26, 106, 126, 136, 170, 171 incenter, 218 index of mathematical terms, 37 infinite descent method, 56, 67 inscribable, 8, 22, 23, 26, 70, 137, 140, 141, 169 integer part, 4, 39 integral equation, 34, 209 integral operator, 211 intrinsic perimeter, 164 isosceles, 18, 21, 25, 122, 134, 135, 138, 159, 160 isotomic point, 218 Jacobian matrix, 190 lattice, 26, 122, 129, 165 Legendre symbol, 72, 82 Lemoine (symmedian) point, 218 length, 19, 21–27, 29, 94, 96, 124, 125, 135, 137, 140, 143, 145, 146, 151, 152, 154–156, 159, 160, 162, 166–169, 171, 175, 182, 187, 211, 217 linear linear combination, 88, 93, 209 linear equation, 113 linear function, 33, 86, 204, 207 linear functional, 131 linear independence, 13, 25, 91, 92, 162 linear map, 131 matrix, 26, 86, 87, 91, 109, 111, 141, 142, 162, 164, 186, 190 median, 27, 175, 217 Mersenne prime number, 55 Mersenne primes, 59 Minkowski norm, 164 mittenpunkt, 218 Nagel line, 219 Nagel point, 23, 141, 218 nilpotent, 90, 93 non-collinear points, 19 noncollinear points, 27, 124, 126, 172 nonlinear operator, 214 number of combinations, 3, 37 orthocenter, 218

Index of Mathematical Terms orthogonal, 130, 136, 137, 143, 145, 153–157 packing, 19, 125, 129 parameterization (of a curve), 41 partition, 15, 17, 26, 99, 103, 113, 114, 151, 153, 156, 158, 168, 169, 171, 174 pentagonal number, 8, 69 perfect number, 59 periodic, 31, 193, 203, 225 poder, 181 polygon, 16, 23, 26, 29, 106, 121, 143, 144, 146, 147, 150, 158, 168, 171, 187 polyhedron, 27, 153, 157, 171–173 polynomial characteristic polynomial, 91 irreducible polynomial, 3, 38, 83, 92, 223 monic polynomial, 34, 214 polynomial with arbitrary coefficients, 11, 13, 86, 91, 93, 211, 214 polynomial with integer coefficients, 6, 9, 12, 40, 51, 61, 82, 89 real polynomial, 12, 29, 30, 33, 86, 87, 185, 188, 206, 208 set of polynomials, 32, 197 symmetric polynomial, 85, 89 projective duality, 172 projective transformation, 138 Pythagorean triple, 5, 42, 44, 66, 67, 69 quadratic irrational number, 220 quadratic residue, 50, 68, 72, 73

229

, 29, 185 relatively prime, 39, 40, 46, 65, 78, 86, 93 Riemann zeta function, 54, 194 root, 34, 61, 83, 89, 91–93, 160, 200, 212 sequence, 4, 6, 9, 12, 14, 17, 18, 30–32, 40, 43, 48, 50, 61, 74, 76, 77, 81, 87, 95, 96, 98, 100, 110, 112, 114, 116, 117, 119, 127, 128, 147, 191, 193, 196, 201, 203, 206, 207, 213 series, 77 simplex, 25, 156, 158, 159, 171 stereographic projection, 41 sum-free set, 17, 114 superperfect number, 60 surface, 26, 103, 154, 166, 174 symmedian, 181 symmetric, 137, 154, 155, 161, 166, 174, 175 symmetry, 51, 88, 97, 106, 133, 136, 137, 160, 166, 170, 172, 205 tiling, 121, 149 torus, 13, 93 transcendental number, 55, 128 triangular number, 8, 69 trilinear coordinates, 138, 141, 219 vector space, 90, 130, 208 width, 23, 145, 152, 153, 155, 166, 167

Index of Topics and Methods

Classical arithmetic functions and ap- Arithmetic progressions: 1.10, 1.26, plications: 1.4, 1.29, 1.34, 1.36, 1.42, 1.40, 2.5, 2.21 1.43, 1.44, 1.45, 1.46, 1.69 Representing integers by algebraic Congruences and divisibility: 1.1, 1.3, functions, Waring problem: 1.31, 1.53, 1.8, 1.11, 1.15, 1.17, 1.22, 1.24, 1.27, 1.61, 1.62, 1.63, 1.64 1.38, 1.48, 1.50, 1.51, 1.67, 1.68, 2.7, 2.45 Rational and algebraic numbers: 1.47, Algebraic number theory: 1.70 2.14, 2.15 Groups: 2.10 Polynomials: 1.2, 1.7, 1.30, 2.2, 2.4, 2.9, 2.12, 2.13, 2.24 Combinatorial identities: 2.33, 2.34 Estimates of arithmetically defined Algebraic equations: 2.3, 2.16 functions: 1.5, 1.6, 1.23, 1.28, 1.37, 1.55, 1.65, 1.66, 2.25, 2.35, 2.47 Linear algebra, matrices, determinants: 2.6, 2.11, 2.17, 2.18, 2.40, 2.43 Symmetric functions: 2.1, 2.8 General Diophantine equations (Pyth- Combinatorial properties of sets of inagorean, exponential): 1.12, 1.14, 1.16, tegers: 2.26, 2.27, 2.28, 2.29, 2.41, 2.44 1.20, 1.25, 1.33, 1.35, 1.39, 1.41, 1.49, 1.52, 1.56, 1.57, 1.60 Sequences of integers: 2.19, 2.42, 2.52, 2.53 Pell’s equation and its applications: 1.13, 1.58, 1.59 Counting problems: 2.23, 3.21 Continued fractions: 1.13 Extremal problems concerning sets of Prime numbers, Chebyshev theorem: integers: 2.36, 2.37, 2.48, 2.49, 2.50, 1.9, 1.18, 1.19, 1.21, 1.32, 1.54, 2.46 2.51

232

Index of Topics and Methods

Ramsey-type problems: 2.20, 2.30, 2.32, 2.38, 2.39, 2.56, 2.57, 2.59, 2.62

Geometric inequalities in the triangle: 3.62, 3.63, 3.64, 3.65, 3.66, 3.67, 3.68, 3.69, 3.70, 3.71, 3.72, 3.73

Graphs and applications: 2.22, 2.31, 2.58 Topological methods in geometry: 3.32, 3.33, 3.34, 3.35, 3.36, 3.38, 3.39, 3.58, Tiling, packing and covering: 2.55, 3.59, 3.61 2.60, 2.64, 3.46, 3.47, 3.53

Area and volume: 3.42, 3.45, 3.51, 3.56, General inequalities: 4.2, 4.3, 4.4, 4.6, 4.14, 4.15, 4.22, 4.26 3.60 Distances in combinatorial geometry: Convexity of real functions: 4.16 2.61, 2.63, 3.23, 3.48, 3.54 Geometry of the triangle: 3.1, 3.11, 3.12, 3.14, 3.16 Complex numbers and applications: 3.5, 3.6, 3.8, 4.1, 4.19, 4.34

Minima and maxima of real functions, mean values: 4.7, 4.8, 4.9, 4.10, 4.37 Functions of a complex variable, holomorphic functions: 4.20, 4.24, 4.44, 4.47

Geometric constructions, geometric loci: 3.2, 3.3, 3.4, 3.7, 3.9, 3.10, 3.15 Sums of series and convergence of sequences: 4.11, 4.12, 4.17, 4.18, 4.23, Trigonometry: 3.17, 3.30 4.28, 4.29, 4.31 Analytic methods in geometry: 3.18 Dynamical systems, density of sets of Higher dimensional geometry: 3.13, reals: 3.19, 4.32, 4.35, 4.43 3.57 Integrals: 4.13, 4.21, 4.30 Convex geometry, integral formulas, isoperimetric inequalities: 3.24, 3.25, Functional equations and inequations: 3.29, 3.49, 3.50, 3.55 4.27, 4.33, 4.36, 4.38 Extremal problems concerning finite sets of points: 2.54, 3.20, 3.27, 3.28, Integral and differential equations: 4.25, 4.39, 4.40, 4.41, 4.42, 4.45 3.31, 3.37, 3.40, 3.41, 3.44, 3.52 General geometric inequalities: 3.22, Linear functionals and operators: 2.65, 3.26, 3.43, 4.5 4.46, 4.48

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close